WO2025117310A1

WO2025117310A1 - Diabetes prognosis in a feline

Info

Publication number: WO2025117310A1
Application number: PCT/US2024/056803
Authority: WO
Inventors: Mathieu MONTOYA; John Flanagan
Original assignee: Mars Inc
Current assignee: Mars Inc
Priority date: 2023-12-01
Filing date: 2024-11-21
Publication date: 2025-06-05
Anticipated expiration: 2026-06-01

Abstract

The present disclosure relates to a computer-implemented method for determining a risk of occurrence of diabetes in a feline including the steps of determining the risk of occurrence of diabetes in the said feline, so as to assess whether the said feline can be prediabetic and determining a risk of occurrence of diabetes within a time interval for the said feline, when determined as being prediabetic. It also pertains to a method for generating a machine learning system suitable for determining the risk of occurrence of diabetes in a feline as well as a computer-implemented system for determining the risk of occurrence of diabetes in a feline.

Description

DIABETES PROGNOSIS IN A FELINE

CROSS-REFERENCE TO RELATED APPLICATIONS

This application claims the benefit of priority of European Patent Application No. 23213725.7, filed December 01, 2023, the content of which is incorporated herein by reference in its entirety, and to which priority is claimed.

FIELD OF THE DISCLOSURE

The present disclosure relates to the field of diabetes prognosis in feline.

BACKGROUND OF THE DISCLOSURE

As it is known, diabetes (i.e. also termed diabetes mellitus) is a condition in which the mammal body cannot properly produce (such as for example in Type I diabetes) or respond (such as for example in Type II diabetes) to the hormone insulin.

Feline with diabetes most commonly suffer from a form of diabetes that resembles to the Type II form of the disease found in humans. It is estimated that between 0.2% and 1% of cats will be finally diagnosed with diabetes during their lifetime.

Typically, clinical signs of diabetes in feline can include weight loss, excessive thirst and urination, and in rare cases feline may experience damage to the nerves and the hind limbs.

Prediabetes is a physiological state preceding diabetes.

In humans, prediabetes is a metabolic state between normal glucose homeostasis and diabetes and is diagnosed by demonstrating impaired glucose tolerance (IGT) and/or impaired fasting glucose (IFG).

In contrast, prediabetes state has not yet been defined in feline, especially in cats, and reference values that are used for determining a prediabetes state in human are mostly irrelevant in feline, such as cats. Illustratively, blood glucose content alone cannot be a valuable marker of diabetes in feline because cats exhibit an elevation of glucose, notably as a response to a stress. Thus, although validated cut points may eventually be useful in veterinary clinical practice to identify feline with altered glucose metabolism and at risk of developing diabetes, these concepts are largely confined to human medical practice. This is why, to date, diabetes is typically late diagnosed in feline, including in cats, only once clinical signs are evident.

Unfortunately, there is no cure available for feline diabetes, which makes diabetes diagnosis in feline a late event, even if an insulin therapy or a dietary therapy can somewhat transiently alleviate symptoms and also, in certain cases, lead to a remission state. However, as it is known in the art, about 25%-30% of cats in remission relapse and require intensive and long-lasting insulin therapy. Further, the majority (76%) of diabetic feline in remission have impaired glucose tolerance and some (19%) have impaired fasting glucose, indicating that these felines did not have normal glucose metabolism or clearance.

In contrast to diabetes, prediabetes state is mostly reversible, such as with an adequate regimen, and hence deserve being diagnosed in feline to avoid the occurrence of the disease.

Thus, there is a need in the art for methods allowing prognosing the likelihood of occurrence of a diabetes in a feline, especially in a cat.

SUMMARY OF THE DISCLOSURE

A first aspect of the present disclosure relates to a computer-implemented method for assessing the risk of a feline to be a prediabetic feline, the method comprising: a) providing a set of at least six physiological features of the feline selected from age, weight, breed, blood glucose content, white blood cell count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content, and blood alkaline phosphatase content; b) generating a feature score predictive of the risk of occurrence of diabetes in a feline from the set of physiological features by operating a first machine learning model trained on the set of physiological features; c) determining, based on the feature score, whether the feline is at risk for occurrence of diabetes; d) designating the feline as a prediabetic feline when the feline is at risk for occurrence of diabetes; and e) optionally, generating a diet for the prediabetic feline. In some embodiments of the computer-implemented method, step a) can include providing (i) a set of one or more features of the said feline selected from age, weight and breed of the said feline and (ii) a set of three or more features measured from a sample previously obtained from the said feline selected from blood glucose content, white blood cells count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content and blood alkaline phosphatase content.

In some embodiments of the computer-implemented method, step a) can include providing a set of features consisting of age, weight, breed, blood glucose content, white blood cells count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content and blood alkaline phosphatase content.

The physiological features of the feline selected from age, weight, breed, blood glucose content, white blood cell count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content, and blood alkaline phosphatase content were found to be statistically significant in relation to diabetic conditions in felines.

In addition, the use of at least six physiological features selected from the above list, has been shown to provide sufficient accuracy for prediction while minimizing the amount of data required. In some embodiments of the computer-implemented method, step a) can further include providing second or more order cross-features calculated from the selected features.

Another aspect of the present disclosure relates to a computer-implemented method for assessing the risk of a prediabetic feline to develop diabetes within a time interval, the method comprising: a’) generating a second feature score predictive of the occurrence of diabetes within a time interval for the prediabetic feline from a second set of at least six physiological features of the feline selected from age, weight, breed, blood glucose content, white blood cell count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content, and blood alkaline phosphatase content by operating a second machine learning model trained on the set of physiological features; b’) determining, based on the second feature score, the risk of occurrence of diabetes within a time interval for the prediabetic feline; c’) optionally, generating a diet for the prediabetic feline based on the time of risk of occurrence of diabetes.

In some embodiments of the computer-implemented method, step a’) can include providing (i) a set of one or more features of the said prediabetic feline selected from age, weight and breed of the said feline and (ii) a set of three or more features measured from a sample previously obtained from the said prediabetic feline selected from blood glucose content, white blood cells count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content and blood alkaline phosphatase content.

In some embodiments of the computer-implemented method, step a’) can include providing a second set of features consisting of age, weight, breed, blood glucose content, white blood cells count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content and blood alkaline phosphatase content.

In some embodiments of the computer-implemented method, step a’) can further include providing second or more order cross-features calculated from the selected features of step e).

In some embodiments of the computer-implemented method, the cross-features can consist of third order cross-features.

In some embodiments of the computer-implemented method, the selected features of steps a) and a’) can be identical.

In some embodiments of the computer-implemented method, each of the first and second machine learning models used at step b) and a’) can be a trained neural network model.

In some embodiments of the computer-implemented method, the trained neural network model can be a trained multilayer perceptron neural network model.

Neural networks models, and more particularly multilayer perceptron neural network models, are well-suited to such predictions.

Predication made by many neural networks may be interpret as Bayesian a posteriori probabilities, see for example M. D Richard, and al.“ Neural Network Classifiers Estimate Bayesian a posteriori Probabilities”. During the training step, the neural networks leam to minimize a risk based on the trained data set, generally following the principle of the gradient descent.

In some embodiments of the computer-implemented method, step b’) can include determining the risk of occurrence of diabetes for a prediabetic feline within a six-month time interval following the time the features of the said feline were measured, preferably within a three-month time interval following the time the features of the said feline were measured.

Another aspect of the present disclosure relates to a computer-implemented method for assessing the risk of a feline to develop diabetes, the method including: a) providing a set of at least six physiological features of the feline selected from age, weight, breed, blood glucose content, white blood cell count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content, and blood alkaline phosphatase content; b) generating a first feature score predictive of the risk of occurrence of diabetes in a feline from the set of physiological features by operating a first machine learning model trained on the set of physiological features; c) determining, based on the first feature score, whether the feline is at risk for occurrence of diabetes; d) designating the feline as a prediabetic feline when the feline is at risk for occurrence of diabetes; e) generating a second feature score predictive of the occurrence of diabetes within a time interval for the prediabetic feline from a second set of at least six physiological features of the feline selected from age, weight, breed, blood glucose content, white blood cell count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content, and blood alkaline phosphatase content, by operating a second machine learning model trained on the second set of physiological features, the second set of physiological features being identical or different from the set of physiological features of step b) ; f) determining, based on the second feature score, the risk of occurrence of diabetes within a time interval for the prediabetic feline; g) optionally, generating a diet for the prediabetic feline based on the time of risk of occurrence of diabetes.

Another aspect of the present disclosure relates to a method for generating a machine learning system suitable for assessing the risk of a feline to be a prediabetic feline including the steps of: a) generating a model suitable for determining the risk of occurrence of diabetes in a feline including the steps of: i) providing a computer-implemented machine learning device, ii) training the said machine learning device by providing a set of data for each of a plurality of felines, wherein the said set of data for each of the plurality of felines includes:

- a set of at least six physiological features of the said feline selected from age, weight, breed, plasma glucose level, white blood cells count (WBC), plasma cholesterol level, plasma phosphorus content, plasma albumin level and plasma alkaline phosphatase content, and

- clinical data relating to the occurrence of diabetes for the said feline, whereby a machine learning model suitable for determining the risk of a feline to be a prediabetic feline is generated.

Another aspect of the present disclosure relates to a method for generating a machine learning system suitable for assessing the risk of occurrence of diabetes within a time interval in a prediabetic feline including the steps of: a’) generating a second model suitable for determining a risk of occurrence of diabetes within a time interval for a prediabetic feline including the steps of : i) providing a second computer-implemented machine learning device, ii) training the said second machine learning device by providing a set of data for each of a plurality of prediabetic felines, wherein the said set of data for each of the plurality of prediabetic felines includes: - a set of at least six of physiological features of the said prediabetic feline selected from age, weight, breed, plasma glucose level, white blood cells count (WBC), plasma cholesterol level, plasma phosphorus content, plasma albumin level and plasma alkaline phosphatase content, and

- clinical data relating to the occurrence of diabetes for the said prediabetic feline, whereby a second machine learning model suitable for determining the risk of occurrence of diabetes within a time interval for a prediabetic feline is generated.

Another aspect of the present disclosure relates to a method for generating a machine learning system suitable for assessing the risk of occurrence of diabetes in a feline including the steps of : a) generating a first model suitable for determining the risk of occurrence of diabetes in a feline including the steps of : i) providing a first computer-implemented machine learning device, ii) training the said first machine learning device by providing a set of data for each of a plurality of felines, wherein the said set of data for each of the plurality of felines includes:

- a set of at least six physiological features of the said feline selected from age, weight, breed, blood glucose content, white blood cells count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content and blood alkaline phosphatase content, and

- clinical data relating to the occurrence of diabetes for the said feline, whereby a first machine learning model suitable for assessing the risk of a feline to be a prediabetic feline is generated, b) generating a second model suitable for assessing a risk of occurrence of diabetes within a time interval for a prediabetic feline including the steps of : i) providing a second computer-implemented machine learning device, ii) training the said second machine learning device by providing a set of data for each of a plurality of prediabetic felines, wherein the said set of data for each of the plurality of prediabetic felines includes: - a set of at least six physiological features of the said prediabetic feline selected from age, weight, breed, blood glucose content, white blood cells count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content and blood alkaline phosphatase content, and

- clinical data relating to the occurrence of diabetes for the said prediabetic feline, whereby a second machine learning model suitable for assessing the risk of occurrence of diabetes within a time interval for a prediabetic feline is generated.

In some embodiments of the said method for generating a machine learning system, step a)-ii) can include training the said first machine learning device with :

- a collection of data sets provided from felines which are labelled as not affected with diabetes, and

- a collection of data sets provided from felines which are labelled as prediabetic.

In some embodiments of the said method for generating a machine learning system, step b)-ii) or a’)-ii) can include training the said second machine learning device with a collection of data sets from felines which are labelled as prediabetic.

Another aspect of the present disclosure relates to a computer-implemented system for determining the risk of a feline to be a prediabetic feline including:

- a processor,

- a tangible computer-readable medium operatively connected to the processor and including a computer code configured to : a) generating a model suitable for determining the risk of a feline to be a prediabetic feline including: i) record a plurality of data sets, each data set being measured for each feline of a plurality of felines, and each data set comprising a plurality of feline physiological features selected from age, weight, breed, plasma glucose level, white blood cells count (WBC), plasma cholesterol level, plasma phosphorus content, plasma albumin level and plasma alkaline phosphatase content, ii) record, for each of the plurality of felines of step i) a diabetes occurrence value, iii) training a machine learning algorithm to generate a machine learning model by using the database to develop a trained model for the risk of occurrence of diabetes, configured to generate feature score wherein the feature score is predictive of the risk of a feline to be a prediabetic feline.

Another aspect of the present disclosure relates to a computer-implemented system for determining the risk of occurrence of diabetes within a time interval in a prediabetic feline including:

- a processor,

- a tangible computer-readable medium operatively connected to the processor and including a computer code configured to: a’) generating a second model suitable for determining a risk of occurrence of diabetes within a time interval for a prediabetic feline comprising: i) record a plurality of data sets, each data set being measured for each feline of a plurality of prediabetic felines, and each data set comprising a plurality of prediabetic feline physiological features selected from age, weight, breed, plasma glucose level, white blood cells count (WBC), plasma cholesterol level, plasma phosphorus content, plasma albumin level and plasma alkaline phosphatase content, ii) record, for each of the plurality of prediabetic felines of step i) a diabetes time occurrence value, iii) training a machine learning algorithm to generate a second machine learning model by using the database to develop a trained model for the risk of occurrence of diabetes within a time interval, configured to generate a feature score, wherein the feature score is predictive of the risk of occurrence of diabetes within a time interval for the said prediabetic feline.

Another aspect of the present disclosure relates to a computer-implemented system for assessing the risk of occurrence of diabetes in a feline including:

- a processor,

- a tangible computer-readable medium operatively connected to the processor and including a computer code configured to : a) acquiring a set of at least six physiological features of the feline selected from age, weight, breed, blood glucose content, white blood cell count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content, and blood alkaline phosphatase content; b) generating a first feature score predictive of the risk of occurrence of diabetes in the feline from the set of physiological features by operating a first machine learning model trained on the set of physiological features; c) determining, based on the first feature score, whether the feline is at risk for occurrence of diabetes; d) designating the feline as a prediabetic feline when the feline is at risk for occurrence of diabetes; e) generating a second feature score predictive of the occurrence of diabetes within a time interval for the prediabetic feline from the set of the physiological features by operating a second machine learning model trained on the set of physiological features; f) determining, based on the second feature score, the risk of occurrence of diabetes within a time interval for the prediabetic feline; g) optionally, generating a diet for the prediabetic feline based on the time of risk of occurrence of diabetes.

In an embodiment, the set of physiological features of step b) is different from the set of physiological features of step e), each comprising at least six physiological features of the feline selected from age, weight, breed, blood glucose content, white blood cell count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content, and blood alkaline phosphatase content, physiological features used at both steps being acquired at step a).

The tangible computer-readable medium can include a computer code configured to: a) generating a first model suitable for assessing the risk of a feline to be a prediabetic feline including: i) record, in a first database, a plurality of data sets, each data set being measured for each feline of a plurality of felines, and each data set including a plurality of feline physiological features selected from age, weight, breed, blood glucose content, white blood cells count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content and blood alkaline phosphatase content, ii) record, in the first database, for each of the plurality of felines of step i) a diabetes occurrence value, iii) training a machine learning algorithm to generate a first machine learning model using the first database to develop an enhanced model for assessing the risk of occurrence of diabetes, configured to generate feature score wherein the feature score is predictive of the risk of a feline to be a prediabetic feline.

In some embodiments, the tangible computer-readable medium can also include, in addition, a computer code configured to: b) generating a second model suitable for assessing the risk of occurrence of diabetes within a time interval for a feline including : i) record, in a second database, a plurality of data sets, each data set being measured for each feline of a plurality of prediabetic felines, and each data set including a plurality of prediabetic feline physiological features selected from age, weight, breed, blood glucose content, white blood cells count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content and blood alkaline phosphatase content, ii) record, in the second database, for each of the plurality of prediabetic felines of step i) a diabetes time occurrence value, iii) training a machine learning algorithm to generate a second machine learning model using the second database to develop an enhanced model for the risk of occurrence of diabetes within a time interval, configured to generate a feature score, wherein the feature score is predictive of the risk of occurrence of diabetes within a time interval for the said prediabetic feline.

First database can be enriched by assessing the risk of a feline to be a prediabetic feline by performing at least steps a) to d) of the computer-implemented method as above described and recording the set of features as a data set in said first database, and the risk of occurrence of a diabetes as a diabetes occurrence value. Second database can be enriched by assessing the risk of occurrence of a diabetes within a time interval for a prediabetic feline by performing at least steps a) to f), or a’) to c’); of the computer-implemented methods as above described; and recording the set of features as a data set in said second database, and the risk of occurrence of a diabetes within a time interval as a diabetes time occurrence value.

The first and the second databases can be the same database.

A further aspect of the present disclosure relates to a method for preventing the occurrence of diabetes in a feline including the steps of:

1) assessing the risk of occurrence of a diabetes in the feline by performing the computer- implemented method according to the specification, and

2) providing an appropriate diet to the feline if a risk of occurrence of a diabetes in the said feline has been determined at step 1).

In some embodiments of the method above, step 2) can further include providing to the said feline a preventive or a dietary regimen.

Another aspect of the present disclosure relates to a method for preventing the occurrence of diabetes within a time interval in a prediabetic feline including the steps of:

1 ’) assessing the risk of occurrence of a diabetes within a time in the said prediabetic feline by performing the computer-implemented method as above described, and

2’) providing a preventing or a therapeutic treatment to the said prediabetic feline if a risk of occurrence of a diabetes within a time interval in the said prediabetic feline has been determined at step 1’).

In some embodiments of the method above, step 2’) can include administering to the said prediabetic feline a preventive or a dietary regimen.

DESCRIPTION OF THE FIGURES

Fig. 1 illustrates an embodiment of a computer-implemented method for assessing the risk of a feline to be a prediabetic feline; Fig. 2 illustrates an embodiment of a computer-implemented method for assessing the risk of a feline to develop diabetes within a time interval;

Fig. 3 illustrates an example of a machine learning model suitable for the implementation of method for assessing the risk of a feline to be a prediabetic feline and/or method for assessing the risk of a feline to develop diabetes within a time interval; and

Fig. 4 illustrates a flow diagram for a computer-implemented method for assessing the risk of a feline to be a prediabetic feline in accordance with an embodiment.

DETAILED DISCLOSURE

It is provided herein a method for assessing a risk of occurrence of a diabetes in a feline, that will be described in detail further in the present description.

In one or more examples, the described techniques may be implemented in hardware, software, firmware, or any combination thereof. If implemented in software, the functions may be stored as one or more instructions or code on a computer-readable medium and executed by a hardware-based processing unit. Computer-readable media may include non- transitory computer-readable media, which corresponds to a tangible medium such as data storage media (e.g., RAM, ROM, EEPROM, flash memory, or any other medium that can be used to store desired program code in the form of instructions or data structures and that can be accessed by a computer).

Instructions may be configurable to be executed by one or more processors, such as one or more digital signal processors (DSPs), general purpose microprocessors, application specific integrated circuits (ASICs), field programmable logic arrays (FPGAs), or other equivalent integrated or discrete logic circuitry. Accordingly, the term “processor” as used herein may refer to any of the foregoing structure or any other physical structure suitable for implementation of the described techniques. Also, the techniques could be fully implemented in one or more circuits or logic elements.

For example, a machine learning-model can be built depending on what selected physiological data has to be provided for a particular subject, then depending on the selected physiological data, a training process may be done to obtain a trained machine learning model. The trained machine learning model can then be invoked at run time based on the selected feline physiological data entered by the user, which feline physiological data can encompass feline contextual information, such as age, weight, breed, and feline physiological data obtained from a sample previously obtained from a tested feline, such as blood glucose content, white blood cells count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content and blood alkaline phosphatase content.

A computer-implemented method as disclosed herein can reside, for example, both within a web-based application as well as a mobile application. The machine learning models can be implemented/deployed at either a backend server that runs the machine learning model as part of a web-based application and communicates information back to the user’s phone or other end device, and/or can also implemented as part of a mobile application that runs on the user’s phone or other end device and communicates information back to the other end device. Regardless of the implementation, the computer-implemented system can send push notifications to the feline’s owner or alternatively to the veterinarian practitioner to alert them in case of an estimated prediabetic state, can receive requests from user, e.g. from the feline’s owner or from the veterinarian practitioner, and deliver precise, coherent answers, and can also provide a pathway for entering feline’s physiological data which can be used to keep track of the feline history as well as improve the prediction estimates.

Definitions

The terms used in this specification generally have their ordinary meanings in the art, within the context of this subject matter and in the specific context where each term is used. Certain terms are defined below to provide additional guidance in describing the compositions and methods of the disclosed subject matter and how to make and use them.

In the following detailed description, reference is made to the accompanying drawings, which form a part hereof. In the drawings, similar symbols typically identify similar components, unless context dictates otherwise. The illustrative embodiments described in the detailed description, drawings, and claims are not meant to be limiting. Other embodiments may be used, and other changes may be made, without departing from the spirit or scope of the subject matter presented here. It will be readily understood that the aspects of the present disclosure, as generally described herein, and illustrated in the figures, can be arranged, substituted, combined, and designed in a wide variety of different configurations, all of which are explicitly contemplated and made part of this disclosure.

In the description herein, references to “embodiments,” “an embodiment,” “one embodiment,” “in various embodiments,” etc., indicate that the embodiment(s) described can include a particular feature, structure, or characteristic, but every embodiment might not necessarily include the particular feature, structure, or characteristic. Moreover, such phrases are not necessarily referring to the same embodiment. Further, when a particular feature, structure, or characteristic is described in connection with an embodiment, it is submitted that it is within the knowledge of one skilled in the art to affect such feature, structure, or characteristic in connection with other embodiments whether or not explicitly described. After reading the description, it will be apparent to one skilled in the relevant art(s) how to implement the disclosure in alternative embodiments.

As used herein, the singular forms “a”, “an”, and “the” include plural referents unless the context clearly dictates otherwise. Thus, for example, reference to “a compound” includes mixtures of compounds.

As used herein, the terms “include”, “including”, “comprises”, “comprising” or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus.

The term “about” or “approximately” means within an acceptable error range for the particular value as determined by one of ordinary skill in the art, which will depend in part on how the value is measured or determined, i.e., the limitations of the measurement system. For example, “about” can mean within three or more than three standard deviations, per the practice in the art. Alternatively, “about” can mean a range of up to 20%, preferably up to 10%, more preferably up to 5%, and more preferably still up to 1% of a given value. Also, particularly with respect to systems or processes, the term can mean within an order of magnitude, preferably within five-fold, and more preferably within two-fold, of a value.

Moreover, the terms "at least", and “ranging from”, encompass the hereafter cited value. For example, "at least 40 ppm" has to be understood as also encompassing "40 ppm". Unless specifically stated otherwise, amounts (in particular weight percentages, amount in parts per million (ppm), or milliequivalents/kg (mEq/kg) fat) are expressed herein by weight of a product or composition reference, for example a preservative food composition according to the disclosure. In the present disclosure, ranges are stated in shorthand, so as to avoid having to set out at length and describe each and every value within the range. Any appropriate value within the range can be selected, where appropriate, as the upper value, lower value, or the terminus of the range. For example, a range from 1 to 10 represents the terminal values of 1 and 10, as well as the intermediate values of 2, 3, 4, 5, 6, 7, 8, 9, and all intermediate ranges encompassed within 1-10, such as 2 to 5, 2 to 8, 7 to 10, etc.

As used herein, the term “blood” encompasses whole blood, blood plasma and blood serum, depending on the context in which it is used. Illustratively, several procedures for determining a selected blood physiological parameter may be performed indifferently in a blood plasma sample or in a blood serum sample, which in all cases lead to a determination value of the said selected physiological parameter from the previously collected blood sample.

As used herein, the term “effective amount” refers to an amount of an ingredient which, when included in a composition, is sufficient to achieve an intended compositional or physiological effect. It is understood that various biological factors can affect the ability of a substance to perform its intended task. Therefore, an "effective amount" can be dependent in some instances on such biological factors. Further, while the achievement of physiological effects can be measured by a skilled person using evaluations known in the art, it is recognized that individual variation can make the achievement of physiological effects a subjective decision. The determination of an effective amount is well within the ordinary skill in the art of nutritional sciences.

As used herein, "administration", and "administering" refer to the manner in which a composition is presented to a subject. Most preferably herein, administration of a food composition can be accomplished by an oral method.

As used herein, "oral administration" refers to a route of administration that can be achieved notably by swallowing or sucking a food composition.

As used herein, the term “food composition” covers all of foodstuff; diet; food; or a material containing at least proteins, carbohydrates and/or fats; which is used in the body of an organism to sustain growth, repair and vital processes; and/or to furnish energy for a companion animal. A food composition as described herein can also contain supplementary substances or additives, for example, minerals, vitamins and condiments (See Merriam-Webster’s Collegiate Dictionary, 10th Edition, 1993). A food composition according to the present disclosure can consist of a nutritionally complete and balanced food composition or a functional complement. The food composition as described herein can be preferably a cooked product.

As used herein, the term “nutritionally complete” refers to animal food products that contain all known required nutrients for the intended recipient of the animal food product, in all appropriate amounts and proportions based, for example, on recommendations of recognized and competent authorities in the field of animal nutrition. Such foods are therefore capable of serving as a source of dietary intake to maintain life, without the addition of supplemental nutritional sources.

As used herein, the term “diabetes” means an incurable chronic disease in which the feline body cannot properly produce (such as for example Type I diabetes) or respond (such as for example Type II diabetes) to the hormone insulin.

As used herein, the term “prediabetes” or sometimes “pre-diabetes” indicates the physiological state, in a feline, and absent any therapeutic intervention (diet, exercise, pharmaceutical, or otherwise) of having a higher than normal expected rate of disease conversion to type 1 or type 2 Diabetes Mellitus. Prediabetes can also refer to those felines who will, or are predicted to convert to, type 1 or type 2 diabetes Mellitus within a given time period or time horizon. It can be stated in terms of a relative risk from normal between quartiles of risk or as a likelihood of diabetes occurrence based on a feature score provided by a computer-implemented method described herein. Unless otherwise noted, and without limitation, when a categorical positive determination of a prediabetic state in a feline is stated according to a computer- implemented method described herein, it can further be defined with a predicted time period before actual conversion to type 1 or type 2 diabetes Mellitus, based on a given threshold value. Thus, a computer-implemented method described herein provides determining the risk of occurrence of diabetes in a feline, i.e. provides determining a prediabetic state in a feline and further encompasses predicting any time period of expected or predicted annual rate of conversion. Felines determined as being in a prediabetic state at steps of a computer- implemented method described herein have thus a predictable risk of conversion to diabetes compared to felines that have not been determined as being prediabetic. In a preferred embodiment, a prediabetic feline is considered as having a risk of occurrence of diabetes within about 12 months. In some embodiments, a prediabetic feline can be considered as having a risk of occurrence of diabetes within about 12 months, about 11 months, about 10 months, about 9 months, about 8 months, about 7 months, about 6 months, about 5 months, about 4 months, about 3 months, about 2 months or about 1 month.

As used herein, the term “time interval” refers to a time interval of about 6 months or less. A time interval shall encompass a period of about 6 months or less, about 5 months or less, about 4 months or less, about 3 months or less, about 2 months or less or about 1 month or less. In some embodiment, a time interval refers to a period of time of about 3 months or less.

As used herein, the term “feline” encompasses animals, including pet animals, selected from cheetah, puma, jaguar, leopard, lion, lynx, liger, tiger, panther, bobcat, ocelot, smilodon, caracal, serval and cats. As used herein, cats encompass wild cats and domestic cats. In particular embodiments, the cats can be domestic cats.

Typically, a feline is labelled as not affected with diabetes at a given time if said feline has not been diagnosed with diabetes within the next 12 months or more from said given time.

Typically, a feline is labelled as prediabetic for the 12 months preceding the date of the diagnosis.

As used herein, the term “risk” relates to the probability that an event will occur, possibly over a specific time period, as in the conversion to frank Diabetes, and can mean a subject's “absolute” risk or “relative” risk. Absolute risk can be measured with reference to either actual observation post-measurement for the relevant time cohort, or with reference to index values developed from statistically valid historical cohorts that have been followed for the relevant time period. Relative risk refers to the ratio of absolute risks of a subject compared either to the absolute risks of low risk cohorts or an average population risk, which can vary by how clinical risk factors are assessed. Odds ratios, the proportion of positive events to negative events for a given test result, are also commonly used (odds are according to the formula p/(l — p) where p is the probability of event and (1— p) is the probability of no event) to no-conversion. Alternative continuous measures which may be assessed in the context of the present invention include time to Diabetes conversion and therapeutic Diabetes conversion risk reduction ratios.

As used herein, “risk evaluation,” or “evaluation of risk” encompasses making a prediction of the probability, odds, or likelihood that an event or disease state may occur, the rate of occurrence of the event or conversion from one disease state to another, i.e., from a non- diabetic condition to a prediabetic condition or prediabetes, or from a prediabetic condition to diabetes. The computer-implemented methods of the present disclosure can be used to make continuous or categorical measurements of the risk of conversion to type 1 or type 2 diabetes, thus diagnosing and defining the risk spectrum of felines determined as being prediabetic. Computer-implemented methods described herein can be used to discriminate between non prediabetic and prediabetic felines. The computer-implemented methods according to the present disclosure can be used so as to discriminate felines according to a predicted time period before the actual occurrence of a diabetic state. Such differing use can require different feline physiological features combinations, mathematical algorithm, and/or threshold points, but be subject to the same aforementioned measurements of accuracy for the intended use.

As used herein, a “physiological feature” of a feline refers to an information related to the said feline, which encompasses contextual information, such as age, weight and breed and a physiological parameter obtained from a sample, such as blood glucose content, white blood cells count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content and blood alkaline phosphatase content. The physiological parameter can be quantitative or qualitative. According to some embodiments, the physiological parameter can be derived from the physiological signal using feature extraction techniques and may include combining a plurality of extracted features and/or parameters, for example by non-linear regression techniques. Within the context of the present disclosure, the terms “feature extraction”, “feature processing” and “signal processing” may refer to the processes, manipulations and signal processing measures performed to analyze a physiological signal in a sample previously obtained from a feline. Non-limiting examples of suitable physiological parameters are depicted in table 1, below.

As used herein, a “sample” is a biological sample isolated from a subject and can include, by way of example and not limitation, whole blood, serum, plasma, blood cells, endothelial cells, tissue biopsies, lymphatic fluid, ascites fluid, interstitial fluid (also known as “extracellular fluid” and encompasses the fluid found in spaces between cells, including, inter aha, gingival crevicular fluid), bone marrow, cerebrospinal fluid (CSF), saliva, mucous, sputum, sweat, urine, or any other secretion, excretion, or other bodily fluids. “Blood sample” refers to whole blood or any fraction thereof, including blood cells, serum and plasma; serum is a preferred blood sample. As used herein, “measuring”, “measure” or “measurement” means assessing the presence, absence, quantity or amount of either a given substance or a given cell type within a feline-derived sample, including the derivation of qualitative or quantitative concentration levels of such substances, or qualitative or quantitative concentration levels of cells.

As used herein, “statistically significant” means that the alteration is greater than what might be expected to happen by chance alone. Statistical significance can be determined by any method known in the art. Commonly used measures of significance include the p-value, which presents the probability of obtaining a result at least as extreme as a given data point, assuming the data point was the result of chance alone. A result is often considered highly significant at a p-value of 0.05 or less.

As used herein, the term “TN” means true negative, which for a disease state test means classifying a non-disease or normal subject correctly.

As used herein, the term “TP” means true positive, which for a disease state test means correctly classifying a disease subject.

As used herein, the term “FN” means false negative, which for a disease state test means classifying a disease subject incorrectly as non-disease or normal.

As used herein, the term “FP” means false positive, which for a disease state test means classifying a normal subject incorrectly as having disease.

As used herein, the term “sensitivity” means a sensitivity value of a prognosis method and is calculated according to the formula TP/(TP+FN) or the true positive fraction of disease subjects.

As used herein, the term “specificity” means a specificity value of a prognosis method and is calculated according to the formula TN/(TN+FP) or the true negative fraction of non-disease or normal subjects.

As used herein, a “trained machine learning model” means that a machine learning model has been provided with labelled data to leam from, that is to say to determine weight and/or bias of the machine learning model to minimize loss. Once trained, a machine learning model may determine from a given input dataset an output. In the disclosure, the given input dataset includes the set of physiological features corresponds to the set of at least six physiological features of the said feline and the output to the feature score being predictive of the risk of occurrence of diabetes in the said feline or the feature score being predictive of the risk of occurrence of diabetes within a time interval for the said prediabetic feline.

As used herein, the term “neural network” refers to various configurations of classifiers used in machine learning, including multilayered perceptrons, with one or more hidden layer, support vector machines and dynamic Bayesian networks. These methods share in common the ability to be trained, the quality of their training evaluated and their ability to make either categorical classifications or of continuous numbers in a regression mode. Typically, a neural network comprises an input layer, an output layer and one or more hidden layers, each hidden layer comprising a plurality of neurons, the number of hidden layer(s) defining the depth of the neural network. A neural network may be fully connected, or not, supervised or unsupervised, may use forward or backward propagation.

As used herein, the terms “computer memory” and “computer memory device” refer to any storage media readable by a computer processor. Examples of computer memory include, but are not limited to, RAM, ROM, computer chips, digital video disc (DVDs), compact discs (CDs), hard disk drives (HDD), and magnetic tape.

As used herein, the term “computer readable medium” refers to any device or system for storing and providing information (e.g., data and instructions) to a computer processor. Examples of computer readable media include, but are not limited to, DVDs, CDs, hard disk drives, magnetic tape and servers for streaming media over networks.

As used herein, the terms “processor” and “central processing unit” or “CPU” are used interchangeably and refer to a device that is able to read a program from a computer memory (e.g., ROM or other computer memory) and perform a set of steps according to the program.

Description

Provided herein is a computer-implemented method for assessing the risk of a feline to be a prediabetic feline. In other words, said computer implemented method allows to assess the risk of occurrence of diabetes within about 12 months or less.

Provided herein is a computer-implemented-method for assessing the risk of occurrence of diabetes within a time interval in a prediabetic feline. In other words, said computer implemented method allows to assess the risk of occurrence of diabetes within about 6 months or less, preferentially within about 3 months or less.

Provided herein is a computer-implemented method for assessing the risk of occurrence of diabetes in a feline including two main steps, a first main step of assessing the risk of occurrence of diabetes in the said feline, which first step is followed by a second main step of assessing a risk of occurrence of diabetes within a time interval, when a risk of occurrence of diabetes has been determined at the end of the first step, that is to say when the said feline has been designated as prediabetic.

Provided herein is a computer-implemented method for assessing the risk of occurrence of diabetes within a time interval in a feline including two main steps, a first main step of assessing the risk of a feline to be a prediabetic feline, which first step can be followed by a second main step of assessing a risk of occurrence of diabetes within a time interval, when a risk of occurrence of diabetes has been determined at the end of the first step, that is to say when the said feline has been designated as prediabetic. According to some embodiments, the first main step of the computer-implemented method including steps a) to d), allows assessing whether the tested feline is likely to be prediabetic, whereas the second main step of the computer-implemented method, comprising steps b’) to c’), allows assessing a time interval starting from the test after which the said feline will be likely to develop diabetes disease.

The methods disclosed herein all stem from the unexpected findings that a plurality of a feline physiological features, when used in combination, are proved to be relevant and indicative of the status of a feline as regards the future occurrence of diabetes.

Unexpectedly, it has been found that the risk of occurrence of diabetes in a feline, thus the occurrence of a prediabetic state in the said feline, can be assessed through a combination of a set of at least six physiological features selected from age, weight, breed, blood glucose content, white blood cells count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content and blood alkaline phosphatase content.

Among these possibly combined physiological features, age, weight and breed are inherent to the feline history and do not need any test method to be assessed.

Blood glucose content of a feline can be determined by any known method such as by following the guidelines on Standard Operating Procedures for Clinical Chemistry published by the World Health Organization in 2020. Blood glucose content can be determined in a blood plasma sample. Typically, determining blood glucose content comprises collecting a blood sample from the said feline, such as in a heparinized container, and determining blood glucose content in the plasma fraction thereof by converting glucose in gluconic acid by using glucose oxidase, which enzyme reaction generates hydrogen peroxide which is subsequently converted in water by the enzyme peroxidase. 4-aminophenazone is used as an oxygen acceptor and the forms with phenol a pink colored chromogen which can be measured at the wavelength of 515 nm. The skilled artisan can also use a commercially available glucometer device, such as a glucometer device compliant with the EN ISO 15197 standard, according to the manufacturer’s instructions.

White blood cell count (WBC) is a conventional physiological feature by which the number of white blood cells in a blood sample is measured and is conventionally reported as part of a standard complete blood count. WBC determination can be performed in a whole blood sample, preferably a heparinized whole blood sample. White blood cell count determination encompasses polynuclear neutrophils, eosinophils, basophils, monocytes, B lymphocytes and T lymphocytes. White blood cell count can be determined by any known method. Illustratively, White blood cell count is determined by a method comprising collecting a blood sample from a feline and determining WBC by using an automated hematology analyzer device, such as for example a Hematlogy Analyzer Coulter LH 750® commercialized by the company Beckman.

Blood cholesterol content is a conventional physiological feature by which the amount of the total cholesterol in a blood sample is measured and is conventionally reported as part of a standard blood analysis. Blood cholesterol content is generally determined from a blood serum sample or form a blood plasma sample. Blood cholesterol content encompasses the additional contents in low density lipoprotein (LDL) and high density lipoprotein (HDL). Blood cholesterol content can be determined by any known method such as by following the guidelines on Standard Operating Procedures for Clinical Chemistry published by the World Health Organization in 2020. Illustratively, the skilled artisan can use the cholesterol test referenced as CHOD-PAP with ATCS commercialized by the company Dialab.

Blood phosphorus content is a conventional physiological feature by which the content in inorganic phosphorus in a blood sample is determined. Blood phosphorus content can be determined from a blood serum sample. Blood phosphorus content can be determined by any known method, such as by following the guidelines on Standard Operating Procedures for Clinical Chemistry published by the World Health Organization in 2020. Illustratively, blood phosphorus content can be determined by a method comprising using, in a blood sample of a feline, ammonium molybdate as a color-forming reagent. Inorganic phosphate contained in the said blood sample forms an ammonium phosphomolybdate complex with ammonium molybdate in the presence of sulfuric acid. The blood phosphorus content in the resulting product is then assessed by measuring the DO value of the resulting sample at 340 nm wavelength (secondary wavelength at 700 nm). The skilled artisan can for example refer to the laboratory procedure and manufacturer’s instructions when using the automated biochemistry analyser commercialized under the name “Cobas 6000 C 501” by the company Roche.

Blood albumin content is a conventional physiological feature by which the content of albumin in a blood sample is determined. Blood albumin content can be determined from a blood plasma sample or from a blood serum sample. Blood albumin content can be determined by any known method, such as by following the guidelines on Standard Operating Procedures for Clinical Chemistry published by the World Health Organization in 2020. Blood albumin content can be determined using the well-known BCG binding method disclosed in the WHO guidelines of 2020. The skilled artisan can for example refer to the laboratory procedure and manufacturer’s instructions when using the automated biochemistry analyser commercialized under the name “Cobas 6000 C 501” by the company Roche.

Blood alkaline phosphatase content is a conventional physiological feature by which the content of alkaline phosphatase is determined in a blood sample. Blood alkaline phosphatase sample can be determined from a blood serum sample or form a heparinized blood plasma sample. Blood alkaline phosphatase content can be determined by any known method, such as by following the guidelines on Standard Operating Procedures for Clinical Chemistry published by the World Health Organization in 2020. Blood alkaline phosphatase content can be determined through a method using paranitrophenyl phosphate, which is colourless, and which is hydrolysed by alkaline phosphatase at 37°C and at pH 10.5 to form free paranitrophenol which is coloured yellow. The addition of NaOH stops the enzyme activity and the final colour shows maximum absorbance at 410 nm. Illustratively, blood alkaline phosphatase content can be determined by using a Unicel DxC 800® Synchron automated analyzer commercialized by the company Beckman Coulter.

Method for assessing the risk of a feline to be a prediabetic feline

The present disclosure relates to a computer-implemented method for assessing the risk of a feline to be a prediabetic feline, the method including: a) providing a set of at least six physiological features of the feline selected from age, weight, breed, blood glucose content, white blood cell count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content, and blood alkaline phosphatase content; b) generating a feature score predictive of the risk of occurrence of diabetes in a feline from the set of physiological features by operating a first machine learning model trained on the set of physiological features; c) determining, based on the feature score, whether the feline is at risk for occurrence of diabetes; d) designating the feline as a prediabetic feline when the feline is at risk for occurrence of diabetes; and e) optionally, generating a diet for the prediabetic feline.

Method for assessing the risk of occurrence of diabetes within a time interval in a prediabetic feline

The present disclosure relates to a computer-implemented method for assessing the risk of a prediabetic feline to develop diabetes within a time interval, the method including: a’) generating a second feature score predictive of the occurrence of diabetes within a time interval for the prediabetic feline from a second set of at least six physiological features of the feline selected from age, weight, breed, blood glucose content, white blood cell count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content, and blood alkaline phosphatase content by operating a second machine learning model trained on the set of physiological features; b’) determining, based on the second feature score, the risk of occurrence of diabetes within a time interval for the prediabetic feline; c’) optionally, generating a diet for the prediabetic feline based on the time of risk of occurrence of diabetes. Method for assessing the risk of occurrence of diabetes in a feline

The present disclosure relates to a computer-implemented method for assessing a risk of occurrence of diabetes in a feline including the steps of : a) providing a set of at least six physiological features of the feline selected from age, weight, breed, blood glucose content, white blood cell count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content, and blood alkaline phosphatase content; b) generating a first feature score predictive of the risk of occurrence of diabetes in a feline from the set of physiological features by operating a first machine learning model trained on the set of physiological features; c) determining, based on the first feature score, whether the feline is at risk for occurrence of diabetes; d) designating the feline as a prediabetic feline when the feline is at risk for occurrence of diabetes; and optionally: e) generating a second feature score predictive of the occurrence of diabetes within a time interval for the prediabetic feline from the set of the physiological features by operating a second machine learning model trained on the set of physiological features; f) determining, based on the second feature score, the risk of occurrence of diabetes within a time interval for the prediabetic feline. g) designating the prediabetic feline as at risk for occurrence of diabetes within a time interval.

The present disclosure relates to a computer-implemented method for assessing a risk of occurrence of diabetes in a feline including the steps of : a) providing a set of at least six physiological features of the feline selected from age, weight, breed, blood glucose content, white blood cell count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content, and blood alkaline phosphatase content; b) generating a first feature score predictive of the risk of occurrence of diabetes in a feline from the set of physiological features by operating a first machine learning model trained on the set of physiological features; c) determining, based on the first feature score, whether the feline is at risk for occurrence of diabetes; d) designating the feline as a prediabetic feline when the feline is at risk for occurrence of diabetes; e) generating a second feature score predictive of the occurrence of diabetes within a time interval for the prediabetic feline from the set of the physiological features by operating a second machine learning model trained on the set of physiological features; f) determining, based on the second feature score, the risk of occurrence of diabetes within a time interval for the prediabetic feline; g) optionally, generating a diet for the prediabetic feline based on the time of risk of occurrence of diabetes.

First main step : steps a) to d) of the method

The first main step) of the method including the steps a) to d) of the method, allows assessing a risk of occurrence of diabetes in a feline. Thus, this first main step allows determining whether a tested feline is likely to be prediabetic.

Steps a) to d) themselves are computer-implemented steps that can be performed by implementing a trained machine learning model that, once the said machine learning model is provided with a set of at least six physiological features of the tested feline, whereby a risk of occurrence of diabetes in the said feline is determined.

In some embodiments, the risk of occurrence of diabetes is materialized, at step b), by a feature score that, independently of its numerical value, is indicative of a probability to develop diabetes in the future.

As it will be detailed in more detail further in the present description, the second main step, including the steps e) to f) of the method is performed only when a risk of occurrence of diabetes in the tested feline has been determined at the end of step d), that is to say only when the tested feline is designated as a prediabetic feline.

Step a) of the method At step a) of the method, a set of at least features of the tested feline is provided. The way the said provided features are actually expressed is not essential. Typically, (i) age and (ii) weight can be expressed as numerical values, e.g. (i) in days, in months or in years or (ii) in grams, hectograms or kilograms. Typically, the breed feature can be expressed as an alphanumeric label, such as the usual name used for the said breed.

In some embodiments of step a), providing a set of at least six features includes providing (i) a set of one or more features of the said feline selected from age, weight and breed of the said feline and (ii) a set of three or more features measured from a sample previously obtained from the said feline selected from blood glucose content, white blood cells count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content and blood alkaline phosphatase content.

According to the method, irrespective of the embodiment thereof which is considered, the provided feline features shall be contemporary, which means that the one or more selected features shall reflect the tested feline at a defined time point. Illustratively, the one or more features selected from blood glucose content, white blood cells count (WBC), blood cholesterol content, blood phosphorus content shall be measured from one or more samples collected at the provided age feature of the feline. Similarly, the one or more features selected from blood glucose content, white blood cells count (WBC), blood cholesterol content, blood phosphorus content shall be measured from the same sample, or alternatively plurality of samples collected at approximately the same time period, e.g. in the same three-month time period, advantageously in the same two-month time period, better in the same one-month time period, such as during the same seven days-time period.

In some embodiments of the method, step a) can include providing the at least six physiological features which are selected in two distinct sets of physiological features, (i) a first set of physiological features that are available by documentation and do not require performing a test, such as age, weight and breed and (ii) a second set of physiological features that are mainly available as a result of a test that is performed on a sample previously obtained from the feline, such as blood glucose content, white blood cells count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content and blood alkaline phosphatase content.

Thus, in some embodiments of the method, step a) can include providing (i) a set of one or more features of the said feline selected from age, weight and breed of the said feline and (ii) a set of three or more features measured from a sample previously obtained from the said feline selected from blood glucose content, white blood cells count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content and blood alkaline phosphatase content.

In some preferred embodiments of the method, step a) can include providing a set of features consisting of age, weight, breed, blood glucose content, white blood cells count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content and blood alkaline phosphatase content.

In some embodiments of the method, step a) can further include providing second or more order cross-features calculated from the selected features. Preferably, the said crossfeatures consist of third order cross-features.

In some embodiments, step a) can include second or more order cross-features and residual blocks calculated from selected features.

A second order cross-feature is a combination of two selected features to generate a new feature. A third order cross feature is a combination of three selected features to generate a new feature.

In one embodiment, all second order cross-features obtained from crossing the selected features in pairs are provided at step a).

In a preferred embodiment, all second order cross-features obtained from crossing the selected features in pairs and all third order cross-features obtained from crossing the selected features three by three are provided at step a).

In some of these embodiments, step a) can include second order cross-features selected from :

- second order cross-features between (i) blood glucose content and (ii) blood phosphorus content leading to blood glucose; blood phosphorus; blood glucose²; blood phosphorus²; blood glucose x blood phosphorus.

In some of these embodiments, step a) can include third order cross-features.

In some preferred embodiments, the selected features of steps a) and e) can be identical. Step b) of the method

At step b) of the method, it is made use of a first trained machine learning model.

The said first trained, i.e. pre-trained, machine learning model can be of various kinds and can include a deep learning architecture. A deep learning architecture comprises a plurality of hidden layers. The more hidden layers a deep learning architecture has, the deeper it is said to be.

Machine learning models, the use of which is encompassed herein, such as neural networks can be designed with a variety of connectivity patterns. In a feed-forward neural network, information is passed from a lower layer to a higher layer, with each neuron in a given layer communicating to a neuron in a higher layer. A neural network can also have a backflow or feedback connection, also known as top-down connection. In a backflow connection, the output from a neuron in a given layer can be communicated to another neuron in the same layer. The reflow architecture can help identify patterns that span more than one input data chunk delivered to the neural network in sequence. The connection from a neuron in a given layer to a neuron in a lower layer is called a feedback connection, that can be also termed a top-down connection. A network with many feedback connections can be helpful when the identification of high-level concepts may assist in discerning particular low-level features of an input.

In particular, deep learning architectures can perform physiological parameters integration tasks by learning to represent input at successively higher levels of abstraction in each layer, thereby building useful feature representations of the input data.

The machine learning model can be trained using the data of a very large population of felines to learn various relationship between the physiological features, and preferably second-order or higher-order cross-features obtained by crossing the physiological features, and the various data categories, and then perform a prediction for a feline based on the feline data relating to the physiological features and the relationships the model has leamt through training.

A machine learning predictive model as disclosed herein can be very powerful in performing a clinical prediction and in assisting a veterinarian practitioner in generating a decision support, e.g. a diet regimen decision support. The feature score can be of various kinds. The feature score can include a probability, or a category chosen from predetermined categories. In an embodiment, the first feature score can include a probability of the risk of occurrence of diabetes and/or a probability of no occurrence of diabetes. The first feature score can be a category chosen from very low risk of occurrence of diabetes, low risk of occurrence of diabetes, medium risk of occurrence, high risk of occurrence of diabetes and very high risk of occurrence of diabetes, or chosen from a set of numerical values representative of a greater or lesser risk of occurrence of diabetes.

In some embodiments, the feature score can be a value expressing the probability of occurrence of diabetes in the said feline.

Step c) of the method

Determining the risk of occurrence of diabetes based on the feature score value that is generated at step b) simply depends on the way the said score value is expressed.

In all cases, for a specific embodiment for the expression of the feature score value, is associated a threshold value that is used as a reference.

Thus, in some embodiments of the feature score, the likely occurrence of diabetes in a tested feline can be predicted when the feature score value is higher than the said threshold value.

In some other embodiments of the feature score, the likely occurrence of diabetes in a tested feline can be predicted when the feature score value is lower than the said threshold value.

Illustratively, in embodiments wherein the feature score expresses the probability of occurrence of diabetes in the said feline, a threshold value can be fixed as being 0.5 and a score value of more than 0.5 is predictive of a likely occurrence of diabetes in the future by the said feline whereas a score value of 0.5 or less is not predictive of a likely occurrence of diabetes in the future by the said feline.

Step d) of the method

When a risk of occurrence of diabetes has not been determined at step c), then the said feline is not categorized as being prediabetic and the method is ended at this method step. When a risk of occurrence of diabetes has been determined at step c), then the tested feline is categorized as being prediabetic, since it is not yet affected with diabetes symptoms but is likely to develop diabetes in the future.

When a risk of occurrence of diabetes has been determined at step c), and the tested feline designated as a prediabetic feline then the method is pursued by performing step e) and f) that are described in more detail below.

Second main step : steps e) to f) of the method

Second main step including steps e) and f) of the method is aimed at assessing, for a feline designated as being prediabetic at step d), the time interval after which the said feline will be likely to actually develop diabetes.

Step e) of the method

At step e) of the method, at least six physiological features of the said prediabetic feline, or measured from a sample previously obtained from the said prediabetic feline, are provided, which are selected from age, weight, breed, blood glucose content, white blood cells count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content and blood alkaline phosphatase content.

In some embodiments of the method, step e) can include providing (i) a set of one or more features of the said prediabetic feline selected from age, weight and breed of the said feline and (ii) a set of three or more features measured from a sample previously obtained from the said prediabetic feline selected from blood glucose content, white blood cells count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content and blood alkaline phosphatase content.

In some preferred embodiments of the method, step e) can include providing a set of features consisting of age, weight, breed, blood glucose content, white blood cells count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content and blood alkaline phosphatase content.

In some other preferred embodiments of the method, step e) can further include providing second or more order cross-features calculated from the selected features of step e). Preferably, the said cross-features can consist of third order cross-features. At step e) of the method, it is made use of a second trained machine learning model.

The said second trained, i.e. pre-trained, machine learning model can be of various kinds and can include a deep learning architecture. The second machine learning model can be of the same kind of the first machine learning model. Typically, the first and the second machine learning model can differ in the number of hidden layers and/or in the number of neurons per layer and/or in the kind of connection between the layers and/or in the input data and/or in the kinds of the feature score as output data.

The feature score can be of various kinds. The second feature score can be of the same kind of the first feature score, or of a different kind from the first feature score. Preferably, the second feature score is a probability, a probability expressing the risk of occurrence of diabetes within a time interval or expression that there is no occurrence of diabetes within a time interval. The second feature score can be a category chosen from very low risk of occurrence of diabetes within a time interval, low risk of occurrence of diabetes within a time interval, medium risk of occurrence within a time interval, high risk of occurrence of diabetes within a time interval and very high risk of occurrence of diabetes within a time interval, or chosen from a set of numerical values representative of a greater or lesser risk of occurrence of diabetes within a time interval.

In some embodiments, the feature score can be a value expressing the probability that diabetes will occur in the said feline after a determined time period.

Step f) of the method

Determining the predicted time interval of occurrence of diabetes based on the feature score value that is generated at step e) simply depends on the way the said score value is expressed.

In all cases, for a specific embodiment for the expression of the feature score value, is associated one or more threshold values that are used as references for each of the plurality of possible time periods after which the said feline is likely to develop diabetes.

Thus, in some embodiments of the feature score, the likely time interval of occurrence of diabetes in the tested feline can be predicted when the feature score value is close to a threshold value indicative of a specific time interval of occurrence of diabetes. In some embodiments of the method, step f) can include assessing the risk of occurrence of diabetes within a six-month time interval, more preferably a three-month time interval for the said feline, following the time the features of the said feline were measured.

Further embodiments of the method

In some embodiments of the method, each of the first and second machine learning models used at step b) and e) can be a trained neural network model.

In embodiments wherein each of the first and second machine learning models used at step b) and e) is a trained neural network model, the said trained neural network model can be a trained multilayer perceptron neural network model.

Method for generating a machine learning system suitable for assessing the risk of a feline to be a prediabetic feline.

The present disclosure further relates to a method for generating a machine learning system suitable for assessing the risk of a feline to be prediabetic feline including the steps of : a) generating a model suitable for determining the risk of occurrence of diabetes in a feline including the steps of: i) providing a computer-implemented machine learning device, ii) training the said machine learning device by providing a set of data for each of a plurality of felines, wherein the said set of data for each of the plurality of felines includes:

In some embodiments of the said method for generating a machine learning system, step a)-ii) can include training the said first machine learning device with : - a collection of data sets provided from felines which are labelled as not affected with diabetes, and

- a collection of data sets provided from felines which are labelled as prediabetics or affected with diabetes.

The machine learning system can be trained with more than two collections of data enabling more than two states to be distinguished.

Method for generating a machine learning system suitable for assessing the risk of occurrence of diabetes within a time interval in a prediabetic feline.

The present disclosure further relates to a method for generating a machine learning system suitable for assessing the risk of occurrence of diabetes within a time interval in a prediabetic feline including the steps of: a’) generating a second model suitable for determining a risk of occurrence of diabetes within a time interval for a prediabetic feline including the steps of : i) providing a second computer-implemented machine learning device, ii) training the said second machine learning device by providing a set of data for each of a plurality of prediabetic felines, wherein the said set of data for each of the plurality of prediabetic felines includes:

- a set of at least six of physiological features of the said prediabetic feline selected from age, weight, breed, plasma glucose level, white blood cells count (WBC), plasma cholesterol level, plasma phosphorus content, plasma albumin level and plasma alkaline phosphatase content, and

In some embodiments of the said method for generating a machine learning system, step a’)-ii) can include training the said second machine learning device with a collection of data sets from felines which are labelled as prediabetic. In some embodiments of the said method for generating a machine learning system, step a’)-ii) can include training the said second machine learning device with :

- a collection of data sets provided from pre-diabetic felines which are labelled as not affected with diabetes for more than a six-month, preferably a three-month, time interval following the data acquisition, and

- a collection of data sets provided from pre-diabetic felines which are labelled as affected with diabetes within a six-month, respectively a three-month, time interval following the data acquisition.

Method for generating a machine learning system suitable for assessing the risk of occurrence of diabetes in a feline.

The present disclosure further relates to a method for generating a machine learning system suitable for assessing the risk of occurrence of diabetes in a feline including the steps of : a) generating a first model suitable for assessing the risk of occurrence of diabetes in a feline including the steps of : i) providing a first computer-implemented machine learning device, ii) training the said first machine learning device by providing a set of data for each of a plurality of felines, wherein the said set of data for each of the plurality of felines includes:

- a set of at least six or more of physiological features of the said feline, or measured from a sample previously obtained from the said feline, selected from age, weight, breed, blood glucose content, white blood cells count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content and blood alkaline phosphatase content, and

- clinical data relating to the occurrence of diabetes for the said feline, whereby a first machine learning model suitable for assessing the risk of occurrence of diabetes in a feline is generated, b) generating a second model suitable for assessing the risk of occurrence of diabetes within a time interval for a feline including the steps of : i) providing a second computer-implemented machine learning device, ii) training the said second machine learning device by providing a set of data for each of a plurality of prediabetic felines, wherein the said set of data for each of the plurality of prediabetic felines includes:

- a set of at least six physiological features of the said prediabetic feline, or measured from a sample previously obtained from the said prediabetic feline, selected from age, weight, breed, blood glucose content, white blood cells count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content and blood alkaline phosphatase content, and

- clinical data relating to the occurrence of diabetes for the said prediabetic feline, whereby a second machine learning model suitable for assessing the risk of occurrence of diabetes within a time interval for a feline is generated.

- a collection of data sets provided from felines which are labelled as prediabetics.

In some embodiments of the said method for generating a machine learning system, step b)-ii) can include training the said second machine learning device with a collection of data sets from felines which are labelled as prediabetic.

In some embodiments of the said method for generating a machine learning system, step b)-ii) can include training the said second machine learning device with :

- a collection of data sets provided from prediabetic felines which are labelled as not affected with diabetes for more than a six-month, preferably a three-month, time interval following the data acquisition, and - a collection of data sets provided from prediabetic felines which are labelled as affected with diabetes within a six-month, respectively a three-month, time interval following the data acquisition.

Computer-implemented system

The present disclosure further relates to a computer-implemented system for determining the risk of a feline to be a prediabetic feline including:

- a processor,

First database can be enriched by assessing the risk of a feline to be a prediabetic feline by performing at least steps a) to d) of the computer-implemented method as above described or a’) to c’); and recording the set of features as a data set in said first database, and the risk of occurrence of a diabetes as a diabetes occurrence value.

The present disclosure further relates to a computer-implemented system for determining the risk of occurrence of diabetes within a time interval in a prediabetic feline including:

- a processor,

The tangible computer-readable medium can include a computer code configured to: a') generating a second model suitable for assessing the risk of occurrence of diabetes within a time interval for a feline including : i) record, in a second database, a plurality of data sets, each data set being measured for each feline of a plurality of prediabetic felines, and each data set including a plurality of prediabetic feline physiological features selected from age, weight, breed, blood glucose content, white blood cells count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content and blood alkaline phosphatase content, ii) record, in the second database, for each of the plurality of prediabetic felines of step i) a diabetes time occurrence value, iii) training a machine learning algorithm to generate a second machine learning model using the second database to develop an enhanced model for the risk of occurrence of diabetes within a time interval, configured to generate a feature score, wherein the feature score is predictive of the risk of occurrence of diabetes within a time interval for the said prediabetic feline.

Second database can be enriched by assessing the risk of occurrence of a diabetes within a time interval for a prediabetic feline by performing at least steps a) to f), or a’) to c’),of the computer-implemented methods as above described and recording the set of features as a data set in said second database, and the risk of occurrence of a diabetes within a time interval as a diabetes time occurrence value. The present disclosure further relates to a computer-implemented system for assessing the risk of occurrence of diabetes in a feline including :

- a processor,

- a tangible computer-readable medium operatively connected to the processor and including a computer code configured to : a) generating a first model suitable for assessing the risk of occurrence of diabetes in a feline including : i) record a plurality of data sets, each data set being measured for each feline of a plurality of felines, and each data set including a plurality of feline physiological features selected from age, weight, breed, blood glucose content, white blood cells count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content and blood alkaline phosphatase content, ii) record, for each of the plurality of felines of step i) a diabetes occurrence value, iii) training a machine learning algorithm to generate a first machine learning model by using the database to develop a trained model for the risk of occurrence of diabetes, configured to generate feature score wherein the feature score is predictive of the risk of occurrence of diabetes in the said feline. b) generating a second model suitable for assessing the risk of occurrence of diabetes within a time interval for a feline including : i) record a plurality of data sets, each data set being measured for each feline of a plurality of prediabetic felines, and each data set including a plurality of prediabetic feline physiological features selected from age, weight, breed, blood glucose content, white blood cells count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content and blood alkaline phosphatase content, ii) record, for each of the plurality of prediabetic felines of step i) a diabetes time occurrence value, iii) training a machine learning algorithm to generate a second machine learning model by using the database to develop a trained model for the risk of occurrence of diabetes within a time interval, configured to generate a feature score, wherein the feature score is predictive of the risk of occurrence of diabetes within a time interval for the said prediabetic feline.

Methods for preventing the occurrence of diabetes in a feline

The present disclosure further relates to a method for preventing the occurrence of diabetes in a feline including the steps of:

The present disclosure further relates to a method for preventing the occurrence of diabetes within a time interval in a prediabetic feline including the steps of:

The present disclosure further relates to a method for preventing the occurrence of diabetes in a feline including the steps of :

1) assessing the risk of occurrence of a diabetes in the said feline by performing the computer- implemented method abode described, and 2) providing an appropriate diet to the said feline if a risk of occurrence of a diabetes in the said feline has been determined at step 1).

In some embodiments of the above methods, step 2) can include administering to the said feline a preventive or a dietary regimen.

According to some embodiments, said preventive or dietary regimen can be specifically formulated to meet their dietary requirements. This dietary regimen can include a diet containing low levels of total sugars (<62g of mono- and di-saccharides per kg of complete feed with a moisture of 12%) as detailed in the European Union Commission Regulation 2020/354 (March 4^th 2020). The dietary regime can otherwise include a diet containing high levels of protein (>40% protein metabolizable energy), as recommended by the American Animal Hospital Association (Behrend et al 2018), such as, in either wet or dry kibble format, Hill’s Prescription Diet m/d, Purina d/m, Purina o/m, Royal Canin Diabetic/Glycobalance, or Royal Canin Satiety. The dietary regime can also include a diet containing low levels of dietary starch (<20%), such as, in either wet or dry kibble format, Hill’s Prescription Diet m/d, Purina d/m, or Royal Canin Diabetic/Glycobalance. Finally, the dietary regimen may include a diet with 2 or more of the previously mentioned attributes.

EXAMPLES

Material & Methods

Figure 1 illustrates the implementation of a method for assessing the risk of occurrence of diabetes in a feline.

Features

In order to predict the risk of occurrence of diabetes in a feline, particular features have to be provided. In particular, as mentioned above, at least six physiological features selected from age, weight, breed, blood glucose content, white blood cell count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content, and blood alkaline phosphatase content have to be provided.

Features may be of different kinds. Categorical features and continuous features may be pre-processed differently. Categorical features may be age or breed whereas weight, blood glucose content, white blood cell count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content, and blood alkaline phosphatase content belongs to continuous features.

Preferably, continuous features may be normalized.

Categorical features may be transformed by means of columns embedding. Transformed categorical features may be normalized.

Second or more order cross features of the features may be calculated, preferably second or more order cross features of normalized features may be calculated.

For example, if we consider six physiological features (fl, f2, f3, f4, f5, f6), the second order cross-features of all the six physiological features, which have been pre-processed, lead to the following features : fl, f2, fi, f4, f5, f6, fl*fl, fl*f2, fl*f3, fl*f4, fl*f5, fl*f6, f2*fl,

f4*f4, f4*f5, f4*f6, f5*fl, f5*f2, f5*f3, f5*f4, f5*f5, f5*f6, f6*fl, f6*f2, f6*f3, f6*f4, f6*f5, f6*f6.

Features selector may be applied to the cross-features.

In a preferred embodiment, the machine learning model comprises a residual block, as illustrated in Figure 3.

The at least six physiological features, preferably after pre-processing, eventually after calculating cross-order features, and optionally applying a features selector, are submitting to a first trained machine learning model.

First trained machine learning model

The first machine learning model can be a neural network, in particular a multilayer perceptron neural network.

A multilayer perceptron neural network comprises at least one input layer, one output layer and one or more hidden layer(s).

Below is detailed an example of a first machine learning that can be used to predict the risk of occurrence of diabetes in a feline. Of course, this example is for illustrative and nonlimiting purposes only, each layers comprising several neurons. Categorical features, such as cat’s breed, are embedded into numerical features that can be understanded by an algorithm. Numerical features, such as blood parameters, are normalized with mean and standard deviation. All features are then concatenated into one fully connected feature vector. From this feature vector of at least 6 features, a new feature vector is generated, consisting of all polynomial combinations of the feature vector with 3^rd order degree, leading to a new feature vector of 83 features. For example, if an input sample is two dimensional and of the form [a, b], the 3^rd order degree-leads to [a, b, a², ab, b², a³, b³, a²b, b²a].

An example of MLP architectures would be composed of one input layer of at least 83 features, feeding a neural network of 4 hidden layers with 8 neurons. Each layer feeds into the next layer and directly into the layers about 2-3 hops away (example of neural network with residual block is disclosed in the article Kaiming He and al. “Deep Residual Learning for Image Recognition”, arXiv.1512.03385, https://doi.org/10.48550/arXiv.1512.03385). As part of the optimization algorithm, the error for the current state of the model is preferably estimated repeatedly. This requires the choice of an error function, conventionally called a loss function, that can be used to estimate the loss of the model so that the weights can be updated to reduce the loss on the next evaluation. Focal loss function has been used to addresses class imbalance during training (see for example the article Tsung-Yi Lin and al., “Focal Loss for Dense Object Detection”, arXiv.1708.02002, https://doi.org/10.48550/arXiv.1708.020Q2). Several other model architecture and loss function could be used leading to similar performance.

The first trained machine learning model outputs a feature score predictive of the risk of occurrence of diabetes in a feline.

Generally, the feature score takes the form of a predication which may be interpreted as Bayesian a posteriori probability.

Determination of a risk of occurrence of diabetes

Based on the feature score provided by the first trained machine learning model, a risk for occurrence of diabetes in a feline is determined.

For example, the feature score may be compared to a threshold value. If the feature score exceeds the threshold value, then there is a risk for occurrence of diabetes and if the feature score is below or equals to the threshold value, then there is no risk, or vice versa. A threshold value may be 0.5, the feature score being comprised in the interval [0;

1].

If it is determined that the feline is at risk of occurrence of diabetes, the feline is labelled as prediabetic.

Figure 2 illustrates the implementation of a method for assessing the risk of a feline to develop diabetes within a time interval.

Contrary to the method for assessing the risk of occurrence of diabetes, such method for assessing the risk of a feline to develop diabetes within a time interval is applied to feline which are categorized as prediabetic.

The feline may be categorized as prediabetic by others means.

Preferably, a method for assessing the risk of occurrence of diabetes is applied to determine if a feline to be tested is at risk and if said feline to be tested is at risk, a method for assessing the risk of said feline to be tested to develop diabetes within a time interval is applied.

Similarly, to the method for assessing the risk of occurrence of diabetes in a feline, the method for assessing the risk of a feline to develop diabetes within a time interval is based on the acquisition, processing, and analysis of particular features.

Notably, as mentioned above, at least six physiological features selected from age, weight, breed, blood glucose content, white blood cell count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content, and blood alkaline phosphatase content have to be provided.

Categorical features and continuous features may be pre-processed differently.

Categorical features may be age or breed whereas weight, blood glucose content, white blood cell count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content, and blood alkaline phosphatase content belongs to continuous features.

Preferably, continuous features may be normalized.

Categorical features may be transformed by means of columns embedding. Transformed categorical features may be normalized. Second or more order cross features of the features may be calculated, preferably second or more order cross features of normalized features may be calculated.

Second trained machine learning model

Categorical features, such as cat’s breed, are embedded into numerical features that can be understanded by an algorithm. Numerical features, such as blood parameters, are normalized with mean and standard deviation. All features are then concatenated into one fully connected feature vector. From this feature vector of at least 6 features, a new feature vector is generated, consisting of all polynomial combinations of the feature vector with 3^rd order degree, leading to a new feature vector of 83 features. For example, if an input sample is two dimensional and of the form [a, b], the 3^rd order degree leads to the followings crossfeatures [a, b, a², ab, b², a³, a²b, b³, b²a].

An example of MLP architectures would be composed of one input layer of at least 83 features, feeding a neural network of 4 hidden layers with 8 neurons. Each layer feeds into the next layer and directly into the layers about 2-3 hops away (neural network with residual block https://doi.org/10.48550/arXiv.1512.03385). As part of the optimization algorithm, the error for the current state of the model must be estimated repeatedly. This requires the choice of an error function, conventionally called a loss function, that can be used to estimate the loss of the model so that the weights can be updated to reduce the loss on the next evaluation. Focal loss function has been used to addresses class imbalance during training (https://doi.org/10.48550/arXiv.1708.020Q2). Several other model architecture and loss function could be used leading to similar performance.

The second trained machine learning model outputs a feature score predictive of the risk for a feline to develop diabetes within a time interval, preferably the time interval being more than two months and less than six months, in particular around three months. Generally, the feature score takes the form of a predication which may be interpreted as Bayesian a posteriori probability.

Determination of a risk for a feline to develop diabetes within a time interval.

Based on the feature score provided by the second trained machine learning model, a risk for a feline to develop diabetes within a time interval is determined.

For example, the feature score may be compared to a threshold value. If the feature score exceeds the threshold value, then there is a risk for the prediabetic feline to be tested to develop diabetes within the time interval and if the feature score is below or equals to the threshold value, then there is no risk, or vice versa.

A threshold value may be 0.5, the feature score being comprised in the interval [0; 1].

Figure 3 schematizes an example of architecture which can be used for the method for assessing the risk of occurrence of diabetes in a feline and/or the method for assessing the risk for a feline to develop diabetes within a time interval.

This example of architecture comprises among others a focal loss function. Such function allows to address the issue of class imbalance problem by adjusting the weight of easy- to-classify cases to classify relative to hard-to-classify cases.

Furthermore, the machine learning model used can be a MLP (multilayer perceptron neural network), which is particularly well-suited to classification issue and especially to nonlinear problems.

Preferably, the architecture used can comprise a residual block. Such block makes it possible to preserve the information from the initial layer to the last layer.

Figure 4 discloses a preferred embodiment of a method 3 to determine whether a feline to be tested is prediabetic or not and, in case such feline to be tested is prediabetic, the risk for said feline to be tested to develop diabetes within a time interval.

Such method 3 comprises the implementation of a method 1 for assessing the risk of occurrence of diabetes according to the invention. If the feline to be tested is designated as prediabetic then a method 2 for assessing the risk for said feline to be tested to develop diabetes within a time interval according to the invention is applied.

Same features can be used for both methods.

In particular, when applied, pre-processing, cross-featuring, and/or features selection can be applied once for both methods.

The features can be assessed from a database where features are stored.

Results

Here below are shown results and associated performances obtained from three machine learning models respectively trained to determine the risk of a feline to be a prediabetic feline (prediabetic model), the risk of a prediabetic feline to develop diabetes within a three-months’ time interval (3 months’ time interval prediabetic model), and the risk of a feline to develop diabetes within a three-months’ time interval (three months’ time interval model). The three machine learning models are multilayer perceptron neural networks comprising one input layer comprising 8 neurons, one output layer comprising 8 neurons, four hidden layers comprising 8 neurons.

Each model was trained with different datasets.

Two clinical datasets issue from two different clinical groups (Clinic 1 and Clinic 2) have been used to train and test the machine learning models, first separately and then by combining both sets. Datasets comprise visits for a plurality of individuals. Depending on the output desired by the machine learning model, and eventually on the input, the visits and individuals selected for the train set and the test set may differ.

Nine trained models are thus evaluated : pre-diabetic model -Clinic 1 ; pre-diabetic model-Clinic 2; pre-diabetic model -Clinic 1 and 2; 3 months’ time interval prediabetic model-Clinic 1; 3 months’ time interval prediabetic model-Clinic 2; 3 months’ time interval prediabetic model- Clinic 1 and 2; three months’ time interval model-Clinic 1; three months’ time interval model- Clinic 2; three months’ time interval model-Clinic 1 and 2. For each dataset, a training set and a test set are created, being respectively based on a 70%/30% split on unique individuals of said dataset. The split on individuals ensures that individuals in one set are not found in another.

The following tables 1 and 2 sum up the distribution of the datasets between train set and test set. Train set and test set comprise individuals labelled as “prediabetic”, the other individuals being labelled as “healthy”. Dataset may comprise several visits for one individual.

Table 1 corresponds to the distribution of datasets Clinic 1, Clinic 2 to perform training and testing phases of pre-diabetic model.

In table 1, an individual is labelled as “pre-diabetes” for the visit(s) 0 to 12 months before a visit during which said individual was detected as being diabetic. Visits of an individual labelled as “pre-diabetes” which are strictly older than 12 months are not considered. Visits of an individual labelled as “pre-diabetes” after the visit during which said individual was detected as being diabetic are not considered either. An individual is labelled as “healthy” if it has not been detected as diabetic during its visits. In table 1, only the visits strictly older than 12 months are considered for healthy individuals, i.e for individuals in which “False” is indicated in the column “pre-diabetes 0-12 months”.

Table 1

Dataset Split Pre-diabetes 0-12 months Number of visits Number of individuals

Clinic 1 train True 8298 7797

False 1254333 388827 test True 3708 3469

False 534412 166522

Clinic 2 train True 3920 259

False 229124 9426 test True 1667 498

False 97109 18865

Table 2 corresponds to the distribution of datasets Clinic 1, Clinic 2 to perform training and testing phases of 3 months’ time interval pre-diabetic model.

In table 2, all individuals are “pre-diabetic”. That is to say, only individuals who have been detected as being diabetic are considered and only the visit(s) 0 to 12 months preceding the visit during which said individual was detected as being diabetic are considered. Among the 0-12 months visits of these pre-diabetic individuals, the “pre-diabetes 0-3 months” column is set to “True” for the 0 to 3 months visit(s) preceding a visit during which a pre-diabetic individual was detected as being diabetic. Visits of pre-diabetic individuals preceding a visit during which the individual was detected as being diabetic of more than 3 months and equal to or less than 12 months are set to “False” for the “pre-diabetes 0-3 months” column. Visits that took place after the visit during which pre-diabetic individuals were detected as being diabetic are not considered.

Table 2

Dataset Split Pre-diabetes 0-3 months Number of visits Number of individuals

Clinic 1 train True 2858 2743

False 32840 13270 test True 733 704

False 8119 3327

Clinic 2 train True 2255 709

False 4753 1160 test True 521 164

False 1154 304

Training phase

The training phase comprises a plurality of epochs during which the training set is used to train the model by defining and updating weights between layers of said model.

A pre-diabetic model has been trained with trained set of Clinic 1 dataset as split in table 1.

A pre-diabetic model has been trained with trained set of Clinic 2 dataset as split in table 1.

A pre-diabetic model has been trained with trained sets of Clinic 1 and Clinic 2 datasets as split in table 1.

A 3 months’ time interval prediabetic model has been trained with trained set of Clinic 1 dataset as split in table 2.

A 3 months’ time interval prediabetic model has been trained with trained set of Clinic 2 dataset as split in table 2. A 3 months’ time interval prediabetic model has been trained with trained sets of Clinic 1 and Clinic 2 datasets as split in table 2.

A three months’ time interval model-Clinic 1 is a combination of the pre-diabetic model trained with trained set of Clinic 1 dataset and the 3 months’ time interval prediabetic model trained with trained set of Clinic 1 dataset.

A three months’ time interval model-Clinic 2 is a combination of the pre-diabetic model trained with trained set of Clinic 2 dataset and the 3 months’ time interval prediabetic model trained with trained set of Clinic 2 dataset.

A three months’ time interval model-Clinic 1 and 2 is a combination of the pre-diabetic model trained with trained set of Clinic 1 and 2 datasets and the 3 months’ time interval prediabetic model trained with trained set of Clinic 1 and 2 datasets.

The combination of a pre-diabetic model and a 3 month’ time interval pre-diabetic model means that both models are implemented, in parallel or preferably successively. If both return “True”, that is to say that the individual is predicted as pre-diabetic and that it risks developing diabetes in a time interval of three-months, then, the three months’ time interval model also returns “True”. Otherwise, the three months’ time interval model returns “False”.

Preferably, when an individual is tested by means of the three months’ time interval model, a pre-diabetic model is first implemented. If the pre-diabetic model returns “False”, the individual is presumed to be healthy and the three months’ time interval model returns “False”. If the pre- diabetic model returns “True”, a 3 months’ time interval prediabetic model is implemented. If the 3 months’ time interval prediabetic model returns “False”, the three months’ time interval model also returns “False”. If the 3 months’ time interval prediabetic model returns “True”, the three months’ time interval model returns “True”.

In a particular embodiment, the three months’ time interval does not return a binary output. For example, if the pre-diabetic model returns “True”, and the 3 months’ time interval prediabetic model returns “False”, the three months’ time interval model may return “Maybe”.

Testing phase

The testing set is not involved in the training phase and is only used to evaluate performance once the model has been trained. The three following tables summarize score metrics representative of each of the nine trained machine learning models performance.

The accuracy refers to the number of correct test results relative to the total number of test results. Mathematically, accuracy is defined as follows: (TP+TN)/(TP+TN+FP+FN), where TP is the number of true positive test results, TN is the number of true negative test results, FP is the number of false positive test results, and FN is the number of false negative test results.

The sensitivity refers to the ability of a model to predict positive test results. Mathematically, sensitivity is defined as follows: TP/(TP+FN)

The specificity refers to the ability of a model to predict negative test results. Mathematically, specificity is defined as follows: TN/(TN+FP)

The precision refers to the portion of correct positive test results relative to the total of positive test results. For example, considering a machine learning model trained to determine whether an individual is prediabetic, the precision corresponds to the number of individuals correctly labelled as “prediabetic” relatively to the total number of individuals labelled as “prediabetic”. Mathematically, precision is defined as follows: (TP)/(TP+FP).

The Fl score is the weighted average of precision and recall. Mathematically, Fl score is defined as follows: (TP)/[TP+ 1/2(FN+FP)]

The prevalence refers to the total number of individuals who have or should have been predicted as positive relative to the number total of individuals. Mathematically, prevalence is defined as follows: (TP+FP)/(TP+TN+FP+FN) pre-diabetic models

Table 3

3 months’ time interval p re-diabetic models

Table 4

3 months’ time interval models (prediabetic model + model 0-3 month) Table 5

The positive predictive values (PPV) and negative predictive values (NPV) relative to the table 5 are shown in the table 6 below:

Table 6

The positive predictive value is the portion of correct positive test results relative to the total of positive test results. Mathematically, positive predictive value is defined as follows: (TP)/(TP+FP). Positive predictive value equals to the precision.

The negative predictive value is the portion of correct negative test results relative to the total of negative test results. Mathematically, the negative predictive value is defined as follows: TN/(TN+FN).

Bibliography references:

[1] M. D Richard, and al.“ Neural Network Classifiers Estimate Bayesian a posteriori Probabilities”

[2] Merriam -Webster’s Collegiate Dictionary, 10^th Edition, 1993

[3] Guidelines on Standard Operating Procedures for Clinical Chemistry published by the World Health Organization in 2020

[3] “AAHA Diabetes Management Guidelines for Dogs and Cats”, Behrend et al 2018, American Animal Hospital Association,

[4] Kaiming He and al. “Deep Residual Learning for Image Recognition”, arXiv.1512.03385, https://doi.org/10.48550/arXiv.1512.03385foo

[5] Tsung-Yi Lin and al., “Focal Loss for Dense Object Detection”, arXiv.1708.02002, https://doi.org/10.48550/arXiv.1708.020Q2

Claims

1. A computer-implemented method for assessing the risk of a feline to be a prediabetic feline, the method comprising: a) providing a set of at least six physiological features of the feline selected from age, weight, breed, blood glucose content, white blood cell count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content, and blood alkaline phosphatase content; b) generating a feature score predictive of the risk of occurrence of diabetes in a feline from the set of physiological features by operating a first machine learning model trained on the set of physiological features; c) determining, based on the feature score, whether the feline is at risk for occurrence of diabetes; d) designating the feline as a prediabetic feline when the feline is at risk for occurrence of diabetes; and e) optionally, generating a diet for the prediabetic feline.

2. A computer-implemented method for assessing the risk of a feline to develop diabetes within a time interval, the method comprising: determining if the feline is a prediabetic feline according to the method of claim 1, a’) generating a second feature score predictive of the occurrence of diabetes within a time interval for the prediabetic feline from a second set of at least six physiological features of the feline selected from age, weight, breed, blood glucose content, white blood cell count (WBC), blood cholesterol content, blood phosphorus content, blood albumin content, and blood alkaline phosphatase content, by operating a second machine learning model trained on the second set of physiological features, the second set of physiological features being identical or different from the set of physiological features of step b) ; b’) determining, based on the second feature score, the risk of occurrence of diabetes within a time interval for the prediabetic feline; c’) optionally, generating a diet for the prediabetic feline based on the time of risk of occurrence of diabetes.

3. The computer-implemented method according to any one of the preceding claims, wherein the said time interval is 6 months, preferentially 3 months.

4. The computer-implemented method according to any one of the preceding claims, wherein step a) comprises providing (i) a set of one or more features of the said feline selected in the group consisting of age, weight and breed of the said feline and (ii) a set of three or more features measured from a sample previously obtained from the said feline selected in the group consisting of plasma glucose level, white blood cells count (WBC), plasma cholesterol level, plasma phosphorus content, plasma albumin level and plasma alkaline phosphatase content.

5. The computer-implemented method according to any one of the preceding claims, wherein step a) comprises providing a set of features consisting of age, weight, breed, plasma glucose level, white blood cells count (WBC), plasma cholesterol level, plasma phosphorus content, plasma albumin level and plasma alkaline phosphatase content.

6. The computer-implemented method according to any one of the preceding claims, wherein step a) further comprises providing second or more order cross-features calculated from the selected features.

7. The computer-implemented method according to any one of claims 2 to 6, wherein step a’) comprises providing (i) a set of one or more features of the said prediabetic feline selected in the group consisting of age, weight and breed of the said feline and (ii) a set of three or more features measured from a sample previously obtained from the said prediabetic feline selected in the group consisting of plasma glucose level, white blood cells count (WBC), plasma cholesterol level, plasma phosphorus content, plasma albumin level and plasma alkaline phosphatase content.

8. The computer-implemented method according to any one of claims 2 to 7, wherein step a’) comprises providing a set of features consisting of age, weight, breed, plasma glucose level, white blood cells count (WBC), plasma cholesterol level, plasma phosphorus content, plasma albumin level and plasma alkaline phosphatase content.

9. The computer-implemented method according to any one of claims 2 to 8, wherein step a’) further comprises providing second or more order cross-features calculated from the selected features.

10. The computer-implemented method according to any one of claims 6 and 9, wherein the cross-features consist of third order cross-features.

11. The computer-implemented method according to any one of claims 1 to 6, and 10, wherein the selected features of steps a) and a’) are identical.

12. The computer-implemented method according to any one of claims 1 to 11, wherein each of the first and second machine learning models used at step b) and b’) is a trained neural network model.

13. The computer-implemented method according to claim 12, wherein the trained neural network model is a trained multilayer perceptron neural network model.

14. A method for generating a machine learning system suitable for assessing the risk of a feline to be a prediabetic feline comprising the steps of : a) generating a model suitable for determining the risk of occurrence of diabetes in a feline comprising the steps of : i) providing a computer-implemented machine learning device, ii) training the said machine learning device by providing a set of data for each of a plurality of felines, wherein the said set of data for each of the plurality of felines comprises:

15. A method for generating a machine learning system suitable for assessing the risk of occurrence of diabetes within a time interval in a prediabetic feline comprising the steps of: a’) generating a first machine learning system suitable for assessing the risk of a feline to be a prediabetic feline according to claim 14: b’) generating a second model suitable for determining a risk of occurrence of diabetes within a time interval for a prediabetic feline comprising the steps of : i) providing a second computer-implemented machine learning device, ii) training the said second machine learning device by providing a set of data for each of a plurality of prediabetic felines, wherein the said set of data for each of the plurality of prediabetic felines comprises:

- a set of at least six physiological features of the said prediabetic feline , selected from age, weight, breed, plasma glucose level, white blood cells count (WBC), plasma cholesterol level, plasma phosphorus content, plasma albumin level and plasma alkaline phosphatase content, and

16. The method according to claim 14; wherein step a)-ii) comprises training the said machine learning device with :

17. The method according to claim 15, wherein step b’)-ii) comprises training the said second machine learning device with a collection of data sets from felines which are labelled as prediabetic.

18. A computer-implemented system for determining the risk of a feline to be a prediabetic feline comprising:

- a processor,

- a tangible computer-readable medium operatively connected to the processor and including a computer code configured to : a) generating a model suitable for determining the risk of a feline to be a prediabetic feline comprising : i) record a plurality of data sets, each data set being measured for each feline of a plurality of felines, and each data set comprising a plurality of feline physiological features selected from age, weight, breed, plasma glucose level, white blood cells count (WBC), plasma cholesterol level, plasma phosphorus content, plasma albumin level and plasma alkaline phosphatase content, ii) record, for each of the plurality of felines of step i) a diabetes occurrence value, iii) training a machine learning algorithm to generate a machine learning model by using the database to develop a trained model for the risk of occurrence of diabetes, configured to generate feature score wherein the feature score is predictive of the risk of a feline to be a prediabetic feline.

19. A computer-implemented system for determining the risk of occurrence of diabetes within a time interval in a prediabetic feline comprising:

- a processor,

- a tangible computer-readable medium operatively connected to the processor and including a computer code configured to: a’) generating a first model suitable for determining the risk of a feline to be a prediabetic feline according to claim 18 b’) generating a second model suitable for determining a risk of occurrence of diabetes within a time interval for a prediabetic feline comprising : i) record a plurality of data sets, each data set being measured for each feline of a plurality of prediabetic felines, and each data set comprising a plurality of prediabetic feline physiological features selected from age, weight, breed, plasma glucose level, white blood cells count (WBC), plasma cholesterol level, plasma phosphorus content, plasma albumin level and plasma alkaline phosphatase content, ii) record, for each of the plurality of prediabetic felines of step i) a diabetes time occurrence value, iii) training a machine learning algorithm so as to generate a second machine learning model by using the database to develop a trained model for the risk of occurrence of diabetes within a time interval, configured to generate a feature score, wherein the feature score is predictive of the risk of occurrence of diabetes within a time interval for the said prediabetic feline.