WO2024209024A1 - Method for customising a biosensor to measure an analyte with high performance parameters by means of artificial intelligence - Google Patents

Abstract

The present invention relates to a method (100) for customising a biosensor (1) to measure a specific analyte, comprising: - determining at least one quality parameter of the biosensor (1), - determining at least one performance parameter and an associated threshold value of the biosensor (1) based on the at least one quality parameter using a trained first artificial intelligence (S2), and - customising the biosensor to measure a specific analyte by evaluating at least one of the performance parameters and the associated threshold value for a customised application scenario (S3).

Description

Method for individualizing a biosensor for measuring an analyte with high performance parameters using artificial intelligence

technical field

The present invention relates to a method for individualizing a biosensor for measuring a specific analyte in order to derive an improved, preferably best possible qualitative and/or quantitative statement about the presence and/or concentration of the analyte(s) in the sample.

State of the art

A highly sensitive method for detecting and analyzing analytes in a sample is to use an analyte-induced change in the surface tension of a miniaturized spring element, a so-called cantilever, to measure the presence and/or concentration of an analyte in a sample.

A sensor device for detecting an occurrence and/or a concentration and/or an amount of an analyte in a sample, comprising a sensor, connection electronics and a housing, wherein the sensor is configured to convert chemical and/or biochemical information of an analyte in a sample into an electrical signal, wherein the sensor comprises a test cantilever having a base and a deformable part, wherein a receptor layer for selectively absorbing the analyte is applied at least to the deformable part, wherein the sensor comprises a reference cantilever having a base and a deformable part, wherein a reference layer for selectively not absorbing the analyte is applied to the deformable part, is known from WO 2022/200438 A1.

US 2022/0335817 A1 discloses a sensor device for detecting an environmental parameter and a method for determining information about a functional state of a sensor device. representation of the invention

In order to further improve the accuracy of the detection of the analyte, a method for individualizing a biosensor for measuring a specific analyte with the features of claim 1 is proposed.

Accordingly, a method for individualizing a biosensor for measuring a specific analyte is proposed, which comprises the steps of determining at least one quality parameter of the biosensor, determining at least one performance parameter and an associated threshold value of the biosensor based on the at least one quality parameter using a trained first artificial intelligence, and individualizing the biosensor for measuring a specific analyte by evaluating at least one of the performance parameters and the associated threshold value for an individual application scenario.

In this way, a biosensor can be customized during or after its manufacture for use in a specific application scenario.

When manufacturing biosensors, for example when manufacturing MEMS biosensors by growing the MEMS structures on a wafer, applying nanosensors or immobilizing the biochemistry, statistical variations or fluctuations in the manufacturing process can lead to different properties of the biosensors. For example, some biosensors may be more suitable for a first application scenario, whereas other biosensors may be more suitable for a second application scenario.

Using the proposed method for individualizing biosensors, they can then be precisely assigned to an application scenario according to their properties, in which the individualized biosensor has a good, preferably optimal, measurement accuracy.

The quality parameter can be determined on the basis of a property of the biosensor, wherein at least one electrical resistance is determined as a property, preferably all electrical resistances of a resistance bridge of the biosensor.

If the biosensor comprises at least one resistive bridge, at least one of the quality parameters may be a self-similarity of the electrical resistances of the resistive bridge. It has been shown that the self-similarity of the resistances of the resistance bridge is a good indicator of the performance of the biosensor and can therefore be used as at least one of the quality parameters that can be used to individualize the respective biosensor.

At least one of the performance parameters of the biosensor may include sensitivity and/or specificity and/or yield. These performance parameters can be used to describe the performance of a biosensor for a specific analyte.

It is of great importance for the user of a biosensor that the respective analysis results are consistent and reliable or that the user receives an indication of the probability or certainty of the respective analysis results.

In other words, the user wants to know whether the test result obtained is reliable or not. However, the expected level of reliability can vary depending on the area of application. For a virus test for humans, a very high level of reliability may be required, for example in order to reliably order or lift quarantine measures. For a home test for the virus as a preliminary test before going to the doctor, or for testing an animal, for example a pet, the expected reliability may also be lower.

In order to quantify the reliability of tests, the parameters of sensitivity and specificity are known, for example, by means of which a test can be classified. For example, in a test for the presence of a certain virus, these parameters can be specified as performance parameters to describe the reliability of the test.

The sensitivity of a test indicates what proportion of the people tested are reliably identified as sick ("positive") by the test. The sensitivity is defined as the number of people tested as sick by the test divided by the number of people who are actually sick in the group of people tested with the test. If the test has a high sensitivity, hardly any sick person will be falsely classified as healthy, so that with a test with a high sensitivity, a sick person will also be identified as sick with a high probability. The specificity of a test indicates which proportion of the people tested are reliably identified as healthy ("negative") by the test. The specificity is defined as the number of people tested as healthy by the test divided by the number of actually healthy people in the group of people tested with the test. If the test is highly specific, hardly any healthy person will be falsely classified as sick, so that with a test with a high specificity, a healthy person will also be identified as healthy with a high probability.

Accordingly, these performance parameters are well suited to determine the performance of the biosensor.

The trained first artificial intelligence, for example a neural network, can determine at least one of the performance parameters and an associated threshold value based on a feature vector, wherein the feature vector is obtained from a data preparation of the at least one quality parameter.

The trained first artificial intelligence can be trained based on training data obtained using the following steps:

Determining a time-resolved measurement curve of at least one specific analyte and at least one sample without the specific analyte, each using a plurality of biosensors;

Determining an individual measurement result and a validity parameter for each biosensor using a second artificial intelligence from the time-resolved measurement curve of the respective biosensor;

Determining at least one quality parameter for each biosensor;

Statistical evaluation of the individual measurement results based on the assigned validity parameters using threshold values, the assigned quality parameters and the previously known result for each biosensor;

Determining performance parameters from statistical evaluation;

Using the quality parameters, performance parameters and thresholds as training data. The time-resolved measurement curves can preferably be recorded at one or different defined and thus known concentrations of the analyte. The training data can also be determined for different specific analytes, for example for different groups of viruses or bacteria.

Preferably, the measurements are recorded using at least one defined measurement routine, for example the measurement routine that is also used when the respective type of biosensor is later used. Different measurement routines, which can differ depending on the application scenario or the measuring equipment used, can also be used for the measurement.

The time-resolved measurement curves can also be recorded in the event of at least one defined disturbance to the measurement, such as a premature termination of the measurement, an excessive change in temperature during the measurement, a displacement of the biosensor during the measurement. Training data can also be generated for different disturbances, such as disturbances due to temperature or interfering proteins, by adapting the samples and/or measurement routine accordingly.

This allows optimized training data to be obtained in order to have a representative data set for training a first trained artificial intelligence, so that the trained first artificial intelligence can in turn carry out a precise individualization of a biosensor.

In particular, the trained first artificial intelligence can be trained on the basis of training data that includes as quality parameters the self-similarity of the electrical resistances of a resistance bridge of each biosensor.

The trained first artificial intelligence may be trained on the basis of training data that includes as performance parameters the sensitivity and/or the specificity and/or the yield of each biosensor.

By training the first artificial intelligence, a particularly good prediction of the suitability of the individual biosensor for a specific application scenario can be made, so that individualization of the biosensor is reliably possible and the measurement of the analyte can be reliably achieved or the economically best application of the biosensor can be determined.

Based on its individualization, the biosensor may undergo no or at least one further manufacturing step, wherein the no or at least one further manufacturing step preferably comprises a storage of quality parameters and/or performance parameters and/or threshold values on the biosensor or a database and/or the at least one further manufacturing step comprises a functionalization of the biosensor, wherein the functionalization of the biosensor preferably comprises an application of a receptor layer and/or reference layer.

The individualization of biosensors makes it possible to assign biosensors to at least one suitable application scenario early in production or at any other time. Furthermore, a biosensor whose individualization does not produce a suitable application scenario and/or whose performance parameters are insufficient or inconsistent can be sorted out early and/or reworked accordingly. This can save manufacturing resources and/or material. And it can be ensured that every biosensor that was assigned to an application scenario in the individualization process is also suitable for this application scenario, preferably as well as possible.

In addition, the individualization of biosensors, which are temporarily stored for example, enables a targeted and rapid continuation and/or reintegration of the individual biosensors into their manufacturing process. The individualization of the biosensors enables an adequate application scenario to be determined for each of the biosensors. Each of the biosensors can then possibly go through at least one further manufacturing step depending on the individual application scenario.

Data from the individualization of the biosensors, such as their performance parameters, application scenario, and/or threshold value, can also be stored in a memory, or stored in the memory before intermediate storage and used to continue and/or reintegrate the individual biosensors into their manufacturing process. The individualization of biosensors thus enables targeted and efficient control of the production and logistics of biosensors, including their inventory. A management system for production and logistics, which is used for can be used to manufacture biosensors, can therefore include and use the process for individualizing biosensors. Biosensors can also be checked again for their performance after transport, if necessary based on a previous individualization. Consequently, biosensors that were damaged during transport can then be sorted out and/or reworked through the inspection.

The process for individualizing biosensors is also suitable as a tool for reusing previously used biosensors to measure a specific analyte. Based on the application of the process for individualizing the previously used biosensor, its performance after use is determined. In the event that the performance of the previously used biosensor is still sufficiently guaranteed, the previously used biosensor can undergo at least one further manufacturing step or neutralization step in order to then be used again to measure a specific analyte.

A method for generating training data for training a trained first artificial intelligence for use in the individualization of biosensors is further proposed, comprising the following steps:

Determining a time-resolved measurement curve of at least one specific analyte, as well as at least one sample without the specific analyte, each using a plurality of biosensors;

Determining an individual measurement result and a validity parameter for each biosensor using a second artificial intelligence from the time-resolved measurement curve of the respective biosensor;

Determining at least one quality parameter for each biosensor;

Statistical evaluation of the individual measurement results based on the assigned

Validity parameters using thresholds, the associated quality parameters and the previously known result for each biosensor;

Determining performance parameters from statistical evaluation; Using the quality parameters, performance parameters and thresholds as training data.

The time-resolved measurement curves can preferably be recorded at one or different defined and thus known concentrations of the analyte. The training data can also be determined for different specific analytes, for example for different groups of viruses or bacteria.

Preferably, the measurements are recorded using at least one defined measurement routine, for example the measurement routine that is also used when the respective type of biosensor is later used. Different measurement routines, which can differ depending on the application scenario or the measuring equipment used, can also be used for the measurement.

The time-resolved measurement curves can also be recorded in the event of at least one defined disturbance to the measurement, such as a premature termination of the measurement, an excessive change in temperature during the measurement, a displacement of the biosensor during the measurement. Training data can also be generated for different disturbances, such as disturbances due to temperature or interfering proteins, by adapting the samples and/or measurement routine accordingly.

This allows optimized training data to be obtained in order to have a representative data set for training a first trained artificial intelligence, so that the trained first artificial intelligence can in turn carry out a precise individualization of a biosensor.

These specific training data subsequently allow the training of a trained first artificial intelligence in such a way that the individualization of biosensors can be carried out reliably.

The quality parameter determined for each biosensor may include the self-similarity of the resistances in a resistance bridge of each biosensor.

Furthermore, a biosensor for measuring a specific analyte is proposed, comprising at least one resistance bridge, wherein the biosensor is activated based on a resistance bridge of a certain quality parameter for a specific application scenario for measuring the specific analyte.

The quality parameter can be the self-similarity of the resistors of the resistance bridge.

The biosensor envisaged here can be a biosensor for converting chemical and/or biochemical information of an analyte into an electrical signal in a sample. The biosensor can comprise a test cantilever having a base and a deformable part, wherein a receptor layer for selective uptake of the analyte is applied at least to the deformable part, and a reference cantilever having a base and a deformable part, wherein a reference layer for selective non-uptake of the analyte is applied to the deformable part.

The base of the test cantilever and/or the base of the reference cantilever can be designed as a rigid base. A rigid base is understood to mean that the respective cantilever, i.e. the test cantilever and/or the reference cantilever, is not or essentially not deformed in relation to the deformable part of the respective cantilever. The rigid base is, for example, connected to a substrate, supported by a substrate or machined out of the substrate. The deformable part of the test cantilever and/or the reference cantilever, on the other hand, is not supported by the substrate, but rather protrudes over an edge of the substrate and is designed to be free.

The deformable part of the test cantilever and/or the reference cantilever can be designed to be deflectable, for example. In this case, a deflection of the respective cantilever can be achieved, for example, around a bending edge formed in a transition region between the base and the deflectable region. The bending edge is, for example, the edge of the substrate along which the cantilever is divided into the base and the deformable part.

However, the deformation of the respective cantilever in its deformable part is not limited to a lifting or lowering deformation; the cantilever can also be deformed in itself, for example arched or wavy or distorted.

A sample is a limited amount of a substance taken from a larger amount of the substance, for example from a reservoir, whereby the composition of the sample is representative of the composition of the substance in the reservoir and accordingly consists of The substance occurrence and substance composition of the sample can be used to determine the corresponding occurrence in the reservoir.

For example, a sample can be a saliva sample, or a blood sample, or lymph, or urine, or sweat, or intertissue fluid, or a swab, in particular a throat swab or a nasal swab or a sinus swab, or removed tissue. A sample includes in particular any type of biological sample, thus in particular also samples from animals.

A sample can also be a non-biological sample, for example a sample of a chemical substance. A gaseous sample is also possible.

An analyte is the substance whose presence in the sample is to be qualitatively and/or quantitatively detected or detected with the biosensor. The analyte can in particular be present directly in the sample, or be dissolved in the sample, or adhere to the sample or part of the sample, in particular a sample particle. The analyte can also enter into a chemical, biological and/or physical interaction with the sample, so that the analyte can only be detected indirectly via a corresponding interaction.

In particular, one sample form can be converted into another sample form so that the analyte, or its presence, can be detected in a simple and reliable manner. For example, a swab can be dissolved in a liquid so that the swab dissolved in the liquid is then the actual sample that is examined for the analyte.

The analyte in the sample can also be chemically pretreated, for example - if the analyte is a virus - by breaking down the virus envelope to access nucleocapsid antigens. Furthermore, the analyte can also be "labeled" by such pretreatment in order to amplify the measurement signal. For example, conjugated antibodies can bind to antigens of the analyte in order to create the greatest possible deformation on the cantilever system.

The sample then contains the chemical information and/or biochemical information about the analyte. The chemical information may include, for example, the type of analyte, the concentration of the analyte, the presence of the analyte, the weight of the analyte, the reactivity of the analyte, the density of the analyte, etc. The biochemical information includes the same properties as the chemical information, but these substances may be identified, for example, by biological processes. In particular, we speak of biochemical information when the analyte has a particular influence on the biological cycle, for example the metabolism or the immune system.

The test cantilever includes a passive and an active test transducer and the reference cantilever includes a passive and active reference transducer.

The chemical and/or biochemical information is converted into an electrical signal. This can mean that an electrical signal can be changed or created by the chemical composition of the analyte. This can, for example, affect the conductivity of a circuit. For example, a first biochemical information can be present when the circuit is conducting and a second biochemical information can be present when the circuit is not conducting or has a reduced conductivity.

In addition to the direct accessibility of information, such as via conductivity, it is also possible to infer biochemical information via a physical and/or chemical process and/or an interaction.

For this purpose, the proposed biosensor comprises a reference and a test cantilever. A cantilever is a spring element that has a base and a deformable part. The base is accordingly an immovable part of the cantilever, which is arranged in particular in a stationary manner on a substrate. The deformable part of the cantilever is arranged on the base and protrudes beyond the base.

In particular, the base and the cantilever can be formed as one piece. In other words, the deformable part of the cantilever is suspended from the base on one side. The deformable part of the cantilever can be bent, deflected and stretched by the deformable part protruding beyond the substrate. The spatial limit from which the cantilever is bendable or the cantilever transitions from the base into the deformable part is called the bending edge. The bending edge is usually an edge of the substrate when the cantilever protrudes beyond the base.

If the cantilever is bent, material stresses and forces arise in or on the material of the cantilever, which can be measured. If such material stress and/or force can be measured, it can be concluded that the cantilever is bent. The purpose of the transducers is to determine or measure the deformation of the cantilevers. The active transducers are arranged on the deformable parts of the cantilevers, whereas the passive transducers are arranged on the bases, for example the bases, of the cantilevers. In particular, the electrical properties of a circuit can be influenced via the transducers.

For example, a deformation of the cantilever can lead to an increase in the resistance of a transducer, for example an active transducer, while no deformation of the cantilever also causes no change in the resistance of the transducer. This can be achieved, for example, by designing the transducers according to the principle of a strain gauge, whereby a deformation of the respective cantilever is expressed in a change in the length of the strain gauge of the transducer applied to it and thus a deformation of the cantilever can be detected directly by a change in the resistance of the strain gauge.

Thus, the chemical and/or biochemical information of the analyte becomes detectable via a deformation of the cantilever, a subsequent registration via a transducer, and finally via a change in an electrical property of a circuit.

The deformation of the cantilever can be induced in a substance-specific manner by means of a suitable coating. For this reason, the reference cantilever has a reference layer for selective non-uptake of the analyte, while the test cantilever has a receptor layer for uptake of the analyte.

A receptor layer is a substance that can interact with the analyte. This in turn means that the receptor layer is chosen specifically for each analyte. Similarly, a reference layer is a substance that cannot interact with the analyte. The reference layer is therefore also chosen specifically for the analyte.

In this case, interaction means that the analyte is in chemical and/or biochemical and/or physical interaction with the receptor layer. In particular, the interaction can consist of a binding of the analyte to the receptor layer. An interaction can also consist of the absorption or adsorption or non-specific adhesion of the analyte to the receptor layer. The receptor and reference layers are preferably chemically identical with regard to possible interference and preferably only differ in their interaction with the analyte. A substance that is not the analyte therefore interacts just as strongly or just as weakly with the receptor layer as with the reference layer.

The selective absorption of the analyte on the test cantilever causes a force to act on the test cantilever through the analyte, so that the test cantilever reacts sensitively to the analyte. Accordingly, the other substances in the sample that are not the analyte only contribute to a background noise in the form of a basic bending of the test cantilever. For example, the force on the test cantilever increases more quickly the greater the concentration of the analyte in the sample or the faster the surface of the cantilever is covered with the analyte. A maximum force possible for the respective formation is reached when the cantilever is completely covered.

The selective non-uptake of the analyte at the reference cantilever, on the other hand, means that no force is exerted by the analyte on the reference cantilever, so that only the substances that are not the analyte contribute to a background noise in the form of a basic deflection of the reference cantilever.

This acting force can cause a deformation in the deformable part of the test cantilever, while the deformable part of the reference cantilever is not bent. The basis for the deflection of the cantilever is the change in surface tension due to the interaction with the analyte. The change in surface tension leads to a stretching or compression of the upper (or lower) surface of the cantilever. The different stretching or compression on the upper and lower sides causes an internal force or material stress in the material, which leads to deformation.

In the proposed biosensor, the measurement procedure is drastically simplified by the selective non-uptake of the analyte by the reference cantilever, since the reference cantilever is not sensitive to the analyte and therefore the analyte does not contribute to the background noise. Only the substances that are not the analyte contribute to the background noise of the reference cantilever. In a sense, the selective non-uptake of the analyte by the reference cantilever can cause the reference cantilever to be exposed to the same turbulence, the same thermal drift and the same influence of all substances that are not the analyte as in a reference sample. However, with the difference that the reference signal is directly in the sample liquid or, if applicable, at the corresponding location in the case of a gaseous sample.

In particular, a reference cantilever with a reference layer and a test cantilever with a receptor layer result in a significantly more specific analysis of the analyte than just a reference cantilever without a receptor layer, since both the reference layer and the receptor layer exhibit a specific interaction or non-interaction with the analyte.

The design of the biosensor with reference cantilever and test cantilever has the advantage that two measurements can be taken in the sample at the same time, whereby the measurement of the reference cantilever can calibrate the measurement of the test cantilever. This reduces environmental influences, such as chemical, thermal, mechanical, electrical and fluidic interference, on the respective measurement, so that the presence of the analyte can be concluded from the comparison of the measurement on the test cantilever and the reference cantilever.

These forces or material stresses, such as strains or compressions, acting on the cantilevers can ultimately be detected by the transducers, whereby different levels of strain or compression result in different levels of stress being detected by the transducers.

Short description of the characters

Preferred further embodiments of the invention are explained in more detail by the following description of the figures.

The following show:

Figure 1 shows a first schematic embodiment of an exemplary biosensor for converting chemical and/or biochemical information;

Figure 2 shows another schematic embodiment of a biosensor;

Figure 3 shows a schematic possibility of connecting the electrodes of a biosensor in the form of a resistance bridge; Figure 4 is a schematic flow chart of a method for customizing an individual fabricated biosensor for measuring a specific analyte;

Figure 5A shows a schematic embodiment for determining voltages and/or electrical resistances at a resistance bridge of a biosensor with a measuring device, here in the form of a schematically shown multimeter;

Figure 5B shows an embodiment for determining the at least one property of the biosensor with a schematically shown camera system;

Figure 6A shows a schematically shown neural network as an embodiment of a trained first artificial intelligence for determining performance parameters and an associated threshold value;

Figure 6B shows a schematic flow chart comprising a second artificial intelligence for generating training data;

Figure 7 is a schematic representation of a training data set comprising the performance parameters sensitivity, specificity and yield as a function of the self-similarity of a resistance bridge and a threshold value;

Figure 8 shows a schematic flow chart as an embodiment for an evaluation of a measurement of a biosensor based on its individualization by means of a second artificial intelligence; and

Figure 9 shows a schematic flow chart of an embodiment for going through no or at least one further manufacturing step of a biosensor based on its individualization.

Detailed description of preferred embodiments

Preferred embodiments are described below with reference to the figures. Identical, similar or equivalent elements in the different figures are provided with identical reference symbols, and a repeated description of these elements is partially omitted in order to avoid redundancies. Figure 1 schematically shows a first embodiment of an exemplary biosensor 1 for converting chemical and/or biochemical information. The biosensor 1 comprises a test cantilever 2, which in turn has a base 20 and a deformable part 22. A passive test transducer 200 is arranged on the base 20, while an active test transducer 220 is arranged on the deformable part 22.

Analogous to the test cantilever 2, the biosensor 1 also has a reference cantilever 3, which in turn has a base 30 with a passive reference transducer 300, as well as a deformable part 32 which has an active reference transducer 320.

The purpose of the transducers is to determine or measure the deformation of the cantilevers. The active transducers are arranged on the deformable parts of the cantilevers, whereas the passive transducers are arranged on the bases, for example the bases, of the cantilevers. In particular, the electrical properties of a circuit can be influenced via the transducers.

For example, a deformation of the cantilever can lead to an increase in the resistance of a transducer, for example an active transducer, while no deformation of the cantilever also causes no change in the resistance of the transducer. This can be achieved, for example, by designing the transducers according to the principle of a strain gauge, whereby a deformation of the respective cantilever is expressed in a change in the length of the strain gauge of the transducer applied to it and thus a deformation of the cantilever can be detected directly by a change in the resistance of the strain gauge.

Thus, the chemical and/or biochemical information of the analyte becomes detectable via a deformation of the cantilever, a subsequent registration via a transducer, and finally via a change in an electrical property of a circuit.

The transducer can also detect a contraction of the surface on which it is arranged. In the embodiments shown, however, the transducers are always arranged on surfaces where expansion is expected.

However, the strain and/or change in surface tension and/or force detected by the transducer may also be a bending force or a shearing force or be caused by a bending force or a shearing force or generally be based on the elastic modulus of the respective cantilevers. In particular, the fastening of the deformable part 22, 32 to the base 20, 30 results in the deformable part 22, 32 aligning itself along a bending curve due to the action of a force which is caused by a change in the surface tension of the test cantilever. The resulting bending curve is given in particular by the geometry, in particular the area moment of inertia of the cantilever, as well as by the mass of the cantilever and the modulus of elasticity. The bending curves can be described, for example, according to the beam theory.

The transducers 200, 220, 300, 320 are each connected to an electronics 4 which is able to record or forward a measurement signal from the transducers 200, 220, 300, 320, while the electronics 4 is also able to supply the transducers 200, 220, 300, 320 with current and/or voltage.

The transducers 300, 320, 200, 220 are preferably contacted via electrodes 401, 402, 403, 404. In particular, the active test transducer 220 is connected to the active reference transducer 320 via the electrode 401. Furthermore, the passive test transducer 200 is connected to the passive reference transducer 300 via the electrode 403. The active test transducer 220 is also connected to the passive test transducer 200 via the electrode 402, whereas the active reference transducer 320 is connected to the passive reference transducer 300 via the electrode 404. This results in a total of four electrodes via which the transducers are electrically contacted with one another.

Electrical contact can be achieved in particular by applying the transducers to existing electrodes so that a conductive connection is created.

The biosensor 1 has the task of indicating the occurrence and/or concentration and/or amount of an analyte 90 in a sample 9.

In Figure 1, the sample 9 is a liquid that was prepared, for example, by taking a swab, in particular a nasal swab or a throat swab from a test subject. It is also conceivable that the sample comprises a liquid on an object, such as a rope or a cotton swab. However, it may also be the case that the sample 9 is saliva or blood or another body fluid. However, it may also be the case that the sample 9 is a gargling fluid that the test subject gargled with. It may also be the case that the sample 9 comes from a tissue sample or from another sample taken. Substance was obtained and/or synthesized from the test subject. The analyte 90 can be dissolved in the sample or be present in an undissolved manner as a suspension, dispersion or emulsion.

However, sample 9 and the analyte of interest 90 can also be of non-biological origin.

In any case, the biosensor 1 is intended to examine the sample 9 with regard to the presence and/or concentration and/or amount of the analyte 90 in the sample 9. For this purpose, a receptor layer 24 with which an analyte 90 can interact is applied to the test cantilever 2. For example, the receptor layer 24 can adsorb or absorb the analyte 90. During adsorption, the analyte 90 would adhere to the surface of the receptor layer 24, while during absorption, the analyte 90 would penetrate into the interior of the receptor layer 24.

If the sample 9 contains a specific analyte 90, this can interact with the receptor layer 24. This can lead to a change in the surface tension of the section of the deformable part 22 of the test cantilever 2 covered with the receptor layer 24. This change in the surface tension can be registered by the active test transducer 220, which in turn is interpreted as a measurement signal in the electronics 4.

The change in the surface tension of the section of the deformable part 22 of the test cantilever 2 covered with the receptor layer 24 can also lead to a deformation of the deformable part 22 of the test cantilever 2. The active test transducer 220 can therefore also register a deformation of the deformable part of the test cantilever 2, which in turn is interpreted as a measurement signal in the electronics 4.

However, the interaction with the sample liquid of the sample 9 can already lead to the registration of a force effect by the active test transducer 220, for example in which only a surface tension of the liquid acts on the deformable part 22 of the test cantilever 2. In this case, the presence of an analyte 90 is therefore not responsible for a corresponding resulting deformation or change in the surface tension.

In order to determine the magnitude of this basic effect of sample 9 on test cantilever 2, reference cantilever 3 is brought into contact with sample 9 at the same time as test cantilever 2. For this purpose, the reference cantilever 3 has a reference layer 34 with which an analyte 90 cannot interact or a reference layer 34 which cannot adsorb or absorb the analyte 90. In this case, an interaction with the analyte 90 is to be avoided in order to enable differentiation from the measurement signal of the test cantilever 2.

Since both the test cantilever 2 and the reference cantilever 3 interact with the sample 9, both cantilevers 2, 3 interact with the sample 9 in a similar manner. The difference here, however, is that the test cantilever 2 can also interact with any analyte 90 that may be present via its reference layer 24. Accordingly, the measurement signals of the active transducers 220, 320 differ if an analyte 90 is present in the sample 9. In the simplest case, the size of the difference in the measurement signals can therefore be used to determine the amount of analyte 90 present in the sample 9.

The deformation of the cantilever can be induced in a substance-specific manner by means of a suitable coating. For this reason, the reference cantilever has a reference layer for selective non-uptake of the analyte, while the test cantilever has a receptor layer for uptake of the analyte.

A receptor layer is a substance that can interact with the analyte. This in turn means that the receptor layer is chosen specifically for each analyte. Similarly, a reference layer is a substance that cannot interact with the analyte. The reference layer is therefore also chosen specifically for the analyte.

In this case, interaction means that the analyte is in chemical and/or biochemical and/or physical interaction with the receptor layer. In particular, the interaction can consist of a binding of the analyte to the receptor layer. An interaction can also consist of the absorption or adsorption or non-specific adhesion of the analyte to the receptor layer.

The receptor and reference layers are preferably chemically identical with regard to possible interference and preferably only differ in their interaction with the analyte. A substance that is not the analyte therefore interacts just as strongly or just as weakly with the receptor layer as with the reference layer. The selective absorption of the analyte on the test cantilever causes a force to act on the test cantilever through the analyte, so that the test cantilever reacts sensitively to the analyte. Accordingly, the other substances in the sample that are not the analyte only contribute to a background noise in the form of a basic bending of the test cantilever. For example, the force on the test cantilever increases more quickly the greater the concentration of the analyte in the sample or the faster the surface of the cantilever is covered with the analyte. A maximum force possible for the respective formation is reached when the cantilever is completely covered.

The selective non-uptake of the analyte at the reference cantilever, on the other hand, means that no force is exerted by the analyte on the reference cantilever, so that only the substances that are not the analyte contribute to a background noise in the form of a basic deflection of the reference cantilever.

This acting force can cause a deformation in the deformable part of the test cantilever, while the deformable part of the reference cantilever is not bent. The basis for the deflection of the cantilever is the change in surface tension due to the interaction with the analyte. The change in surface tension leads to a stretching or compression of the upper (or lower) surface of the cantilever. The different stretching or compression on the upper and lower sides causes an internal force or material stress in the material, which leads to deformation.

State-of-the-art reference cantilevers simply do not have a receptor layer that reacts sensitively to the analyte. This means that effects such as turbulence in the sample and the thermal drift of the sensor system can be determined. However, with such a reference cantilever, the analyte can bind to the reference layer of the reference cantilever, for example, through non-specific binding. This means that the analyte itself contributes to the background noise. Therefore, with a state-of-the-art sensor, reference measurements in a reference sample, i.e. a sample without analyte, are necessary. This is the only way to determine the effect of non-specific binding of substances that are not the analyte.

In the biosensor, the measurement process is drastically simplified by the selective non-uptake of the analyte by the reference cantilever, since the reference cantilever is not sensitive to the analyte and therefore the analyte does not contribute to the background noise. Only the Substances that are not the analyte contribute to the background noise of the reference cantilever. To a certain extent, the selective non-uptake of the analyte at the reference cantilever can cause the reference cantilever to be exposed to the same turbulence, the same thermal drift and the same influence of all substances that are not the analyte as in a reference sample. However, with the difference that the reference signal is determined directly in the sample liquid.

In particular, a reference cantilever with a reference layer and a test cantilever with a receptor layer result in a significantly more specific analysis of the analyte than just a reference cantilever without a receptor layer, since both the reference layer and the receptor layer exhibit a specific interaction or non-interaction with the analyte.

The design of the sensor with reference cantilever and test cantilever has the advantage that two measurements can be taken in the sample at the same time, whereby the measurement of the reference cantilever can calibrate the measurement of the test cantilever. This reduces environmental influences, such as chemical, thermal, mechanical, electrical and fluidic interference, on the respective measurement, so that the presence of the analyte can be concluded from the comparison of the measurement on the test cantilever and the reference cantilever.

These forces or material stresses, such as strains or compressions, acting on the cantilevers can ultimately be detected by the transducers, whereby different levels of strain or compression result in different levels of stress being detected by the transducers.

The test cantilever 2 and the reference cantilever 3 measure the presence of the analyte 90 in the sample 9 at different positions. Different sample properties can occur at different positions, such as temperature fluctuations or concentration gradients. These different environmental conditions can be measured with the passive transducers 200, 300.

The passive transducers 200, 300 are arranged on the base and preferably do not detect a measurement signal in the event of a deformation or change in the surface tension of the deformable part 22, 32 of the reference or test cantilevers 2, 3. However, the base level of the measurement signal of the passive transducers 200, 300 can be influenced due to these different environmental conditions. By providing a comparison value for each measured value of the active transducers 220, 230 via the passive transducers 200, 300, which considers the environmental conditions in isolation, the influence of the environmental conditions on the measurement signals of the active transducers 220, 320 can be determined and reduced or eliminated or isolated.

Accordingly, the presence of an analyte 90 in a sample 9 can be analyzed in isolation using the biosensor 1 by reducing and isolating the influence of interactions that cannot be assigned to the analyte 90 using a large number of measuring points on the reference and test cantilevers 3, 2. This enables a high degree of measurement accuracy of the presence of the analyte 90 in the sample 9.

Figure 2 shows a further schematic embodiment of a biosensor 1. The transducers 300, 320, 200, 220 are contacted via the electrodes 401, 402, 403, 404. In particular, the active test transducer 220 is connected to the active reference transducer 320 via the electrode 401. Furthermore, the passive test transducer 200 is connected to the passive reference transducer 300 via the electrode 403. The active test transducer 220 is also connected to the passive test transducer 200 via the electrode 402, whereas the active reference transducer 320 is connected to the passive reference transducer 300 via the electrode 404.

This results in a total of four electrodes via which the transducers 200, 220, 300, 320 are electrically contacted with each other. Electrical contact can be achieved in particular by applying the transducers to the electrodes so that a conductive connection is created.

Figure 3 shows one way of connecting the electrodes. The electrodes that contact the transducers 200, 220, 300, 320 are constructed in mirror symmetry. Currents or voltages run through the electrodes, so that an asymmetrical design of these electrodes could lead to asymmetrical crosstalk of electrical signals to the other electrodes. This mutual influence can lead to the generation of a stray signal between the electrodes, but this can be avoided by the symmetrical design. Voltages therefore drop across the electrical resistors of the transducers 200, 220, 300, 320, for example a single-ended voltage on the passive test transducer Utest passive (200), as well as a single-ended voltage on the passive reference transducer Ureterence passive poo).

The transducers 200, 220, 300, 320 are electrically connected in a so-called resistance bridge. In the resistance bridge, an excitation voltage VO, for example direct voltage or alternating voltage, is applied between the electrodes 403, 401. Between these electrodes, the passive and active transducers act as voltage dividers due to their electrical resistances. A resistance bridge in the form shown has the advantage that no voltage is built up between the electrodes 402, 404, provided that the ratio of the electrical resistances of the passive transducer 200 to the active transducer 220 of the test cantilever 2 is equal to the ratio of the electrical resistances of the passive transducer 300 to the active transducer 320 of the reference cantilever 3. In particular, the deviation of an electrical resistance is sufficient to change the electrical resistance ratios and thus to build up a voltage between the electrodes 402, 404.

When the reference cantilever 3 and the test cantilever 2 interact with the sample 9 and the analyte 90, both deformable parts 22, 32 experience, for example, a change in the surface tension, which is, for example, greater for the deformable part 22 of the test cantilever 2 than for the deformable part 32 of the reference cantilever 3. Consequently, the electrical resistance of the active test transducer 220 of the deformable part 22 of the test cantilever 2 will vary to a greater extent than for the active reference transducer 320 of the deformable part 32 of the reference cantilever 3. If the electrical resistances of the passive transducers 200, 300 do not change or at least change in the same way, a change in the electrical resistance ratios results from the deformation of the deformable part 22 of the test cantilever 2 due to the interaction with the analyte 90 of the sample 9, which is specifically related to the reference layer 24 of the test cantilever 2. In such an interaction, a voltage is accordingly built up between the electrodes 402, 404, so that a change in the surface tension in the active test transducer 220 relative to the active reference transducer 320 can be displayed as a bridge transverse voltage VB. Preferably, the bridge transverse voltage VB scales with the occurrence of the analyte 90 in the sample 9, so that a quantitative evaluation of the measurement signal is possible.

The bridge transverse voltage VB corresponds to the difference between the single-ended voltage at the passive test transducer Utest passive (200) and the single-ended voltage at the passive reference transducer Ureterence passive pooj. A bridge transverse voltage detector 44 can display or forward the bridge transverse voltage VB to the outside, so that it is visible to the user of the biosensor 1 that a bridge transverse voltage VB is present. In particular, such a bridge transverse voltage detector 44 can also be provided by an AD converter 440, wherein the AD converter 440 converts the bridge transverse voltage VB into a digital signal, which can be forwarded to the external measuring device.

In particular, the AD converter 440 can be operated in two different measurement modes.

The first measurement mode is the differential measurement mode in which the bridge transverse voltage VB is measured and thus a relative measurement value for the deformation of the two reference and test cantilevers 3, 2 is generated. In this differential measurement mode, the measurement signal of all transducers 200, 220, 300, 320 is taken into account, so to speak, so that the output signal of the AD converter 440 is a measurement signal cleaned of environmental influences, through which the relative deformation of the deformable parts 32, 22 and thus the presence of an analyte 90 can be deduced.

The second measuring mode is the so-called absolute measuring mode. In the absolute measuring mode, the bridge transverse voltage VB is not detected, but rather the signals at the electrodes 402 and 404 are tapped in isolation from one another, so that a statement can be made about the respective deflections of the deformable parts 32, 22. This information is denied to the user in the differential measuring mode, in which the bridge transverse voltage VB is measured.

The biosensors actually manufactured can have a certain range of variation with regard to the structures applied in each case and thus in particular also with regard to the properties of the transducers, in particular the active and passive test transducers 220, 200 and the active and passive reference transducers 320, 300. These differences in the properties, for example in the electrical resistances of the transducers 200, 220, 300, 320, are due to manufacturing and therefore cannot be completely ruled out.

However, it is of great importance for the user of a biosensor 1 that the respective analysis results are consistent and reliable or that the user receives an indication of the probability or certainty of the respective analysis results. In other words, the user wants to know whether the test result obtained is reliable or not. However, the expected level of reliability can vary depending on the area of application. For a virus test for humans, a very high level of reliability may be required, for example, in order to reliably order or lift quarantine measures. For a lactate measurement for a recreational athlete, the expected reliability may also be lower.

In order to quantify the reliability of tests, the parameters of sensitivity and specificity are known, for example, by means of which a test can be classified. For example, in a test for the presence of a certain virus, these performance parameters can be specified to describe the reliability of the test.

The sensitivity of a test indicates which proportion of the people tested are reliably identified as sick ("positive") by the test. The sensitivity is defined as the number of people tested as sick by the test divided by the number of people who are actually sick in the group of people tested with the test. If the test has a high sensitivity, hardly any sick person will be falsely classified as healthy, so that with a test with a high sensitivity, a sick person will also be identified as sick with a high probability.

The specificity of a test indicates which proportion of the people tested are reliably identified as healthy ("negative") by the test. The specificity is defined as the number of people tested as healthy by the test divided by the number of actually healthy people in the group of people tested with the test. If the test is highly specific, hardly any healthy person will be falsely classified as sick, so that with a test with a high specificity, a healthy person will also be identified as healthy with a high probability.

In the following, a method is described by means of which the biosensors described above can be individualized to measure a specific analyte and the measurement method is then designed in such a way that the test results always enable a measurement result with a high sensitivity and a high specificity, despite a possible range of fluctuations in the production of the biosensors.

Figure 4 shows a schematic flow chart of an embodiment of a method 100 for individualizing an individual, manufactured biosensor 1 for measuring a specific analyte 90. The individualization of the individually manufactured biosensor 1 should be carried out in such a way that the individual biosensor 1 can be used by the user for a specific application scenario after successful individualization. In the corresponding application scenario, it is then provided that the biosensor 1 carries out the corresponding analysis in such a way that the requirements for the reliability of the test expected by the user for the specific application scenario, for example the sensitivity and specificity, can always be met, or the test is rejected. In other words, a reliable test result can always be achieved in this way.

In the method 100, in a first step S1, at least one quality parameter of the individual biosensor 1, the individualization of which is to be achieved, is determined. The at least one quality parameter can be used as an indicator of the performance of the biosensor 1. At least the at least one quality parameter serves in the further method 100 for individualizing the biosensor.

The at least one quality parameter of the individual biosensor 1 results, for example, from the production data of the biosensor 1 that was recorded during the production of the biosensor 1 or from a measurement. The quality parameters can either be determined directly from the production data or the measurement data, or can be determined from the production data or the measurement data. An exemplary determination of measurement data or a determination of quality parameters from the measurement data or from production data is described below, for example in Figures 5A and 5B. Reference is made to this at this point.

In a second step S2, at least one performance parameter and a corresponding threshold value are then determined for the individual biosensor 1 on the basis of the quality parameters provided. The threshold value preferably corresponds to a probability based on which the correctness of a measurement result is assessed by a measurement with the biosensor. The correctness can assume the values to be correct or invalid. The corresponding threshold value depends on the performance parameters of the biosensor 1. The performance parameters that can be determined are, for example, the sensitivity, specificity and/or the yield of the respective biosensor 1 for the measurement of a specific analyte 90. The sensitivity or specificity indicates how reliably a medical diagnostic method, such as the measurement of a specific analyte 90 in a user, detects whether the user is ill or not ill. The yield in turn indicates the extent to which a medical diagnostic procedure, such as measuring a specific analyte in a user, provides a correct result.

The performance parameters are determined by means of a trained first artificial intelligence, whereby the trained first artificial intelligence determines the performance parameters and threshold values of the respective individual biosensor 1 from the quality parameters. After the input of at least one quality parameter, the trained first artificial intelligence will output a list of performance parameters and associated threshold values.

For example, the determination in step S2 results in a list of tuples with performance parameters and associated threshold values that is individual for the respective biosensor 1, based on the at least one quality parameter of the respective biosensor 1. Each tuple from the list of tuples includes a value and an associated threshold value for each performance parameter. If the performance parameters are selected appropriately, the tuple allows an assessment of the correctness of a measurement of the biosensor 1 based on the associated threshold value.

As shown in the embodiment 6A, a neural network 80 can be used as the trained first artificial intelligence for determining performance parameters. The training of the first artificial intelligence is described below following Figure 6A. Reference is made to this at this point.

By training the first artificial intelligence, this is specified for the task in step S2. The first artificial intelligence used in step S2 can thus determine lists for one or more performance parameters with the respective threshold values for each, or a specific, analyte. For example, the sensitivity, specificity and/or the yield of the respective biosensor 1 for the measurement of a specific analyte 90 can be determined as performance parameters by the trained first artificial intelligence.

In a third step S3, the individual biosensor 1 is then individualized on the basis of the determined performance parameters and the associated threshold value in such a way that the performance parameters and their threshold values are analyzed for different application scenarios and in particular analytes 90 to determine whether the individual Biosensor 1 is suitable for a specific application scenario, or rather for another application scenario.

For example, biosensor 1 is to be used in a first application scenario to detect a virus to contain a pandemic. In this case, analyte 90 is clearly specified, because this is the respective virus that occurs in the sample fluid of sample 9 to be tested. Furthermore, in order to be applicable for this case, a high sensitivity and a high specificity of the test must be achieved. For this first application scenario, a requirement profile arises in that analyte 90 in the form of the virus must be determined with a high specificity and at the same time a high sensitivity, and taking the yield into account. A high sensitivity and a high specificity are of great importance here, among other things, in order to be able to contain the pandemic on the one hand, but also to be able to reliably implement official measures to restrict individual freedom, such as quarantine measures, on the other. The test usually has both a specificity and a sensitivity of 100%. A high yield is important for reasons of resource conservation and user satisfaction in order to provide enough tests whose accuracy in measuring the specific analyte is not invalid, for example.

In a second example application scenario, for example, the determination of the lactate content in the blood of a recreational athlete may be required. Here, too, the analyte 90 is clearly specified, namely lactate in the athlete's blood. However, the requirements for the sensitivity and specificity of the corresponding biosensor are not so high in this case, for example, because although the recreational athlete needs the data for individual training control, this does not entail any substantial restrictions on personal freedom, for example through quarantine measures. Accordingly, in this second application scenario, the analyte would be clear, namely lactate in the blood, but the requirements for specificity and sensitivity are significantly lower.

Specifically, in step S3, for example, based on the performance parameters determined in step S2 and the associated threshold values, it can be determined whether an individual biosensor 1 is suitable, for example, for a virus test to contain a pandemic, which requires high sensitivity and high specificity, or whether the individual biosensor 1 is more suitable for an application scenario in which high specificity and high sensitivity are not required. If a list of tuples is available, a tuple can be selected from the list of tuples. The selection of the tuple can preferably be made based on the values for the performance parameters in the list of tuples, taking into account the application scenario, in order to individualize the biosensor.

The threshold value in the selected tuple can later be used when the biosensor measures the specific analyte to assess the correctness of the measurement. The threshold value depends on the performance parameters. This is also discussed in the description of Figure 7 below. Reference is also made to this. Accordingly, the tuple and in particular the threshold value for the respective performance parameter can also advantageously be stored with the biosensor 1.

After completion of the individualization, the process 100 ends and the individual, now customized biosensor can be marked and used according to the respective defined application scenario.

For this purpose, both the application scenario and the associated threshold value can be stored in the respective individual, now customized biosensor.

Of course, with an appropriate evaluation of the individual biosensor, several different application scenarios with the corresponding threshold values can be determined and saved, which can then be selected and stored in the biosensor.

The storage of the data, i.e. the corresponding application scenario and the associated performance parameters and threshold values, can be stored in a non-volatile memory in the biosensor, for example an EEPROM. By storing the data, the individualization of the respective individual biosensor can be completed.

The individualization of the biosensor for measuring a specific analyte can also take place in step S3 using the trained first artificial intelligence from step S2, taking into account a specific application scenario. The trained first artificial intelligence therefore determines the performance parameter values and the corresponding threshold value for the biosensor 1 on the basis of the at least one quality parameter, taking into account the specific application scenario. The trained first artificial intelligence selects, for example, a tuple from a list of tuples based on the specific application scenarios. The performance parameter values and the associated threshold value in the selected tuple are based on the specific application scenario and the trained first artificial intelligence individualizes the biosensor 1. The specific application scenario can, for example, be specified by a user or determined based on current inventory levels and demand and provided to the trained first artificial intelligence.

If no suitable tuple can be selected by the trained first artificial intelligence, for example because no performance parameter values exist for the biosensor 1 for the specific application scenario, the first artificial intelligence can issue an error message. In this case, the trained artificial intelligence can also independently select another specific application scenario, for example based on the current inventory levels and/or demand. The first artificial intelligence takes the specific application scenario into account in order to determine a performance parameter with an associated threshold value for the biosensor 1. Accordingly, the training of the first artificial intelligence takes the specific application scenario into account. The consideration of the specific application scenario when training the first artificial intelligence is explained in more detail at the end of Figure 6B. Reference is made to this at this point.

The determination of at least one quality parameter by measuring measured values or evaluating production data is shown below in Figures 5A and 5B.

Figure 5A schematically shows an embodiment for determining a property of the biosensor 1 based on an embodiment of a biosensor 1 comprising a test cantilever 2 and a reference cantilever 3 with deformable parts 22, 32. By means of a measuring device, shown here as an example in the form of a multimeter 50, the voltages at the electrical resistances of the transducers 300, 320, 200, 220, such as the single-ended voltage at the passive test transducer Utest passive (200) and the single-ended voltage at the passive reference transducer Ureference passive (3oo>, are measured. The transducers 300, 320, 200, 220 are contacted via the electrodes 401, 402, 403, 404. For this purpose, measuring tips 52 of the multimeter 50 are correspondingly connected to the electrodes of the biosensor 1. The measured value of the voltage at the electrical resistance is shown in the embodiment on a display of the multimeter 50. The multimeter 50 can also be used to measure other properties of the biosensor 1, such as electrical Resistances, capacitances or impedances. For example, the multimeter can be used to measure the electrical resistances of the transducers 300, 320, 200, 220. The multimeter 50 can also be part of a measuring apparatus, in particular during or after the manufacturing process of the biosensor, so that the measurement of at least one property can be automated.

Figure 5B shows a further embodiment for determining the at least one property of the subcomponents or all subcomponents of the biosensor 1 using a camera system 60. The camera system 60 can record an image section of the biosensor 1 using an imaging method, create a spectrum of the image section, evaluate a time-resolved analysis, a spectrally filtered excitation and spectrally filtered detection of signals using a camera and, for example, using a computer 62. The evaluation can take place in all production steps and may possibly include the use of electrons, light, ions and/or near-field analyses.

The evaluation of the image section can, for example, result in a determination of the electrical resistance of the transducers 300, 320, 200, 220 as a property of the biosensor 1. The transducers 300, 320, 200, 220 are contacted via the electrodes 401, 402, 403, 404. The evaluation of the image section can also result in the coverage density and homogeneity of a coating of a test cantilever 2 and a reference cantilever 3 as a property of the biosensor 1. The evaluation of the image section can also be the individual, finely resolved, geometric configuration of the cantilever, the electrodes or the transducer and can also contain spectroscopic information.

At least one quality parameter can then be determined from the previously determined properties of the biosensor.

For example, a proportion and its distribution of the functionalized surface of the biosensor 1 can be determined as a quality parameter.

From the previously determined single-ended voltages at the electrical resistances of the passive test transducer Utest passive (200), the passive reference transducer Ureference passive (300) and the excitation voltage V0, the self-similarity of the electrical resistances in the resistance bridge of the biosensor 1 can also be determined as a quality parameter. The self-similarity of the biosensor 1 results here in a first consideration, for example, on the basis of the following calculation rule, taking into account the excitation voltage VO at the resistance bridge, the single-ended voltage at the passive test transducer Utest passive (200) and the single-ended voltage at the passive reference transducer Ureterence passive poo) as follows:

Self-similarity ⁼ min(1 - | Utest passive (200) / (VO / 2) - 11 , 1 - | Ureference passive (300) / (VO / 2) - 11), where min( , ) outputs the smallest value of the arguments. The single-ended voltages at the electrical resistances of the passive test transducer Utest passive (200) and the passive reference transducer Ureterence passive poo) can be determined from the electrical resistances Rreterence passive (2oo), Rreterence active (220), Rtest passive poo), Rtest active (320) of the transducers 200, 220, 300, 320 in the resistance bridge and the excitation voltage VO at the resistance bridge as follows:

Utest passive (200) ⁼ VO * Rtest passive (200) /( Rtest passive (200) + Rtest active (220)) ,

Ureference passive (300) ⁼ VO * Rreterence passive (300) /( Rreterence passive (300) + Rreterence active (320)).

This results in the self-similarity depending on the electrical resistances Rreterence passive (2oo), Rreterence active (220), Rtest passive (300), Rtest active (320) of the transducers 200, 220, 300, 320 as follows:

Self-similarity = min(1 - I (Rtest passive <200) - Rtest active (220)) / (Rtest passive (200) + Rtest active (220)) | , 1 - | (Rreterence passive (300) - Rreterence active (320)) / (Rreterence passive (300) + Rreterence sactive (320)) |) ,

The self-similarity can possibly also be determined by taking into account the bridge transverse stress VB.

As already described above, the trained first artificial intelligence used in step S2 can be a trained neural network 80. In Figure 6A, the trained neural network 80 from step S2 is shown again as an embodiment as a possibility for a first artificial intelligence. The trained first artificial intelligence, for example the neural network 80 from Figure 6A, is trained with specific training data in order to enable the determination of the performance parameters from the quality parameters for the respective biosensor 1, wherein the generation and application of the training data will also be discussed below, based on which the first artificial intelligence can be trained for use in step S2. In step S2, the neural network 80 shown in Figure 6A determines the performance parameters of the biosensor 1 and associated threshold values for an analyte 90 using a feature vector. The feature vector results from the at least one determined quality parameter of the biosensor 1.

The feature vector can be obtained, for example, from data processing of the at least one quality parameter. The data processing can, for example, include rescaling the at least one quality parameter. The data processing can also include an error analysis of the at least one quality parameter, wherein the error analysis ends with the generation of the feature vector. The data processing can include the removal of unnecessary quality parameters. A principal component analysis or another method for data reduction or data selection can also be used during data processing in order to obtain the feature vector that includes as much information as possible of the underlying at least one quality parameter. The data processing can also include storing the at least one quality parameter and/or the feature vector in a database or in a memory, such as an EEPROM, of the biosensor 1. The feature vector can also correspond to the at least one quality parameter.

The neural network 80 shown as an example in the embodiment in Figure 6A has four layers. A neural network 80 can also have a topology with, for example, more or fewer layers, as well as more or fewer neurons in the respective layers.

The feature vector based on the quality parameters is passed as an input value to the trained neural network 80. For example, the feature vector can have n components x = (xi,... ,xn), whereby the neural network 80 accordingly has n neurons 82 in a first layer 84. The neural network 80 in the embodiment in Figure 6A has three neurons 82 in its first layer 84. Feature vectors with three components are thus passed to the neural network 80. A neuron 82 at the i-th position in the second layer 85 of the neural network 80 outputs a value as follows Oi (w, ^T + bi), where w = (wi,... ,w _n ) is a vector with weights and bi is a bias value. Oi is an activation function. The activation function can be, for example, a ReLU or sigmoid function.

The neural network 80 outputs performance parameters and their associated threshold values as output values at its last layer 86. In the embodiment in Figure 6A, the neural network 80 has five neurons 82 in its last layer 86, wherein the five neurons 82 output a performance parameter list 70. A performance parameter list can, for example, be a list of tuples, wherein each tuple from the list of tuples includes a value for each performance parameter and an associated threshold value.

The five neurons 82 can, for example, output five values for the performance parameter specificity in the form of a performance parameter list, each with an associated threshold value. For example, the specificity of the biosensor 1 is selected from the performance parameter list with the five values for the specificity and their associated threshold values.

The trained neural network 80 in Figure 6A is trained by training data 88 and is capable of fulfilling the function described in S2.

Figure 6B shows schematically how the training data 88 can be obtained.

A measurement is carried out under controlled conditions using a large number of biosensors 1 that have already been manufactured. In other words, a biosensor 1 is used to measure a known analyte concentration of the analyte 90 of interest under regular measurement conditions, i.e. the measurement conditions for which the respective biosensor 1 is intended for the user, and under regular environmental influences that are to be expected during a measurement. Likewise, a large number of measurements can also be carried out under controlled disturbances in order to create a more global topography. The course of the resulting time-resolved measurement is stored for each biosensor 1 of the manufactured biosensors in combination with the information about the measured analyte 90 - in particular the known analyte concentration and/or the presence or absence of the analyte 90 in the sample liquid of the sample 9 to be analyzed.

This results in a time-resolved measurement curve 800 for each individual biosensor 1 from the multitude of biosensors already manufactured, as shown as an example for a biosensor in Figure 6B. The measurement accordingly provides a measurement curve 800, wherein the measurement curve is based on a voltage signal at the resistance bridge of the biosensor 1 that varies over time.

It is also known which analyte 90 was measured in which concentration and under which disturbance, so that both the individual measurement curve 800 for each biosensor 1 and the expected result is known. In this way, training data can be provided for training a second artificial intelligence 820.

In the second artificial intelligence 820, for example a classifier, the individual measurement result 822 is determined from the time-resolved measurement curve 800 of an individual measurement and at the same time a validity parameter 824 is output.

For example, the second artificial intelligence 820 can be a classifier that classifies the measurement result into a positive or negative individual measurement result. For this purpose, the second artificial intelligence 820 can, for example, comprise dynamic time normalization (DTW for short, “dynamic time warping”). The second artificial intelligence 820 can also comprise a regression model. The regression model can output an analyte concentration as an individual measurement result.

The validity parameter 824 is compared with a predetermined threshold value to indicate the correctness of the measurement result. The threshold value stored together with the application scenario when customizing the biosensor may be used as a predetermined threshold value for evaluating the validity parameter when evaluating the measurement. The correctness may take the values correct or invalid.

The present measurement result may, for example, be invalid if a time-resolved measurement curve 800 ends abruptly due to a premature termination of the associated measurement and an unknown course of the measurement curve is present. In this case, the second artificial intelligence 820 can determine the individual measurement result for the measurement curve, but only with very low validity. Due to the comparison of the associated validity parameter with the specified threshold value, the associated measurement result may then be classified as invalid.

At the same time, before each measurement of a specific analyte with a biosensor 1, its quality parameters are also determined, for example the self-similarity of the electrical resistances in the resistance bridge.

In a sorting step 840, the determined data, consisting of the individual measurement result 822, the associated validity parameter 824, the associated quality parameter, for example the self-similarity 826, and the known result for each biosensor from the multitude of biosensors already manufactured, are statistically evaluated. The statistical evaluation can include the following sub-steps. The sorting step 840 can initially include filtering the determined data. When filtering the determined data, each assigned validity parameter is compared with a specified threshold value in order to indicate the correctness of each individual measurement result. All individual measurement results whose correctness is, for example, correct are combined with the assigned validity parameters, as well as their assigned quality parameters and the expected results to form a filtered data set. The filtered data set is in turn divided into groups, with each group including biosensors with the same quality parameters. A group can possibly also include all associated data from the filtered data set of which one or more quality parameters are above one or more predetermined limit values. The performance parameters, such as sensitivity, specificity and/or yield, can now be determined for each of these groups. The filtering and subsequent determination of the performance parameters is repeated for a large number of specified threshold values in order to generate the training data 88.

In Figure 6B, the training data 88 are given as an example by a three-dimensional representation, for example for the sensitivity, wherein the values for the sensitivity are displayed on the basis of a grayscale coding 72. The three-dimensional representation has the self-similarity plotted on the y-axis as a quality parameter and the threshold value plotted on the x-axis.

The resulting training data 88, which, when applied as training data to the trained first artificial intelligence, such as the neural network 80 in Figure 6A, result in a list of performance parameters 70, for example comprising sensitivity, specificity and/or yield, with the associated threshold values for the biosensor for the specific analyte 90, being able to be determined when a quality parameter for a biosensor 1, for example self-similarity, is entered.

The training data can also be marked according to the specific application scenarios. For example, the training data can be used to specify that performance parameters within a range are suitable for a specific application, such as lactate measurement or the measurement of a specific virus. If the training data is marked according to the specific application scenarios, the corresponding training data can be used to train an initial artificial intelligence that can individualize the Biosensor for measuring a specific analyte in step S3, as shown for example in Figure 1, including a specific application scenario for the biosensor 1.

The second artificial intelligence 820 can also be used to evaluate a measurement by a user (see Figure 8).

The second artificial intelligence 820 is trained to output the individual measurement result and validity parameter. For training, stored measurement curves of the large number of biosensors already manufactured can be used in combination with the information about the measured analyte - in particular the known analyte concentration and/or the presence or absence of the analyte in the sample liquid to be analyzed. This data is stored in a training data set.

If the second artificial intelligence 820 is, for example, a classifier, the classifier can be based on a nearest neighbor classification method. The classifier determines the measurement result, for example, for an unknown time-resolved measurement curve 800 of a biosensor based on a comparison of the unknown measurement curve with the measurement curves in the training data set.

In the case of measurement curves of equal length and not distorted, the comparison can be carried out, for example, by calculating a norm between the still unknown measurement curve of the biosensor and all measurement curves in the training data set. The magnitude values are then ordered in ascending order. The measurement result of the unknown measurement curve of the biosensor can be obtained by a calculation rule, for example by averaging, of a predefined number of measurement results for measurement curves from the training data set with the smallest magnitude values. The validity parameter is calculated, for example, on the basis of the highest magnitude value for the measurement curve in the training data set, the measurement result of which is used to determine the measurement result of the still unknown measurement curve.

The second artificial intelligence 820 can be adapted to include dynamic time warping (DTW for short). DTW is based on a method that compares two time series, such as two time-resolved measurement curves 800, whereby the two time series do not have to match perfectly, but can be shifted or distorted in time, for example. The DTW enables, for example, two time series of different lengths to be compared with each other. Consequently, DTW can also be used to compare two time-resolved measurement curves 800, which do not run over the same period of time.

If the second artificial intelligence 820 comprises, for example, a regression model, the regression model is trained by minimizing a cost function. The validity parameter can result from a calculation taking into account the measurement result and the training data set.

Figure 7 shows the training data 88 again in three representations comprising sensitivity, specificity and yield.

The values for the performance parameters sensitivity, specificity and yield are shown using a grayscale bar 72. The self-similarity in the three representations lies between 0 and 1, and the threshold value between 0.5 and 0.9.

The three representations show that there is a systematic relationship between the sensitivity, specificity and yield depending on the self-similarity and the threshold, and between the performance parameters sensitivity, specificity and yield for a certain self-similarity and a given threshold. The threshold therefore depends on the performance parameters. The threshold also depends on the self-similarity.

The systematic relationship is evident in the fact that at a certain level of self-similarity and with increasing threshold values, there is a tendency for greater sensitivity and specificity, although the yield tends to decrease.

By including the predetermined threshold value, the sensitivity, specificity and yield for the biosensor 1 can then be determined during the measurement on the basis of the respective performance parameter lists 70. For the determined self-similarity of the biosensor 1, a coordination between the sensitivity and the specificity, as well as the yield of the biosensor 1, can thus be carried out by setting the threshold value.

Figure 8 shows, using a flow chart, an embodiment for an evaluation of a measurement of a biosensor 1 based on the individualization of the biosensor 1. For this purpose, in the method 100, as already described above, the individualized biosensor is used for the application scenario for which it was individualized to measure the corresponding specific analyte.

In step S4, the measurement is then carried out with the appropriately customized biosensor under the specified measurement conditions. The measurement results in the time-resolved measurement curve 800, which shows the course of the signal voltage at the resistance bridge for the reference cantilever 3 and the test cantilever 2 in a time-resolved manner.

The measurement of the specific analyte 90 can be carried out using an apparatus. The apparatus also includes, among other things, the sample 9 to be measured with the specific analyte 90. An apparatus can possibly also include several technical components. The technical components can be, for example, one or more mobile devices. The apparatus can also include one or more programs. The one or more programs can, for example, be stored and/or run on the technical components. The programs can be used to evaluate the measurement of the specific analyte, including the individualized biosensor. The evaluation can, for example, be triggered by an electrical signal from the biosensor 1 to the apparatus. The measurement can also be evaluated on a server based on data from the measurement of the analyte. The measurement data, as well as the data from the biosensors themselves, such as at least one of their quality parameters, application scenarios, performance parameters, and/or threshold values, can also be stored and processed on a cloud.

The resulting measurement curve 800 is fed in step S5 to the second artificial intelligence 820, which outputs an individual measurement result as well as a validity parameter that the measurement result is correct.

Depending on the validity parameter, the final measurement result is then output to the user in step S6, such that in a case where the validity parameter falls below a predetermined threshold, the measurement result is classified as invalid. In cases where the validity parameter is above the predetermined threshold, however, a measurement result can be output that is positive or negative with a corresponding probability that corresponds at least to the validity parameter. The threshold value stored together with the application scenario during the individualization of the biosensor 1 may be used as a predetermined threshold value for the evaluation of the Validity parameter is used in the evaluation of the measurement. The threshold value from the individualization of the biosensor 1 preferably results, among other things, from the application scenario of the biosensor 1.

The comparison between the validity parameter and the specified threshold value can also take place in step S5 by the second artificial intelligence. Step S6 therefore includes an output of the classification of the measurement result, the measurement result itself and/or the corresponding probability, which corresponds at least to the validity parameter.

This means that the measurement result can always be produced absolutely reliably depending on the given application scenario by individualizing the biosensor on the one hand and evaluating the measurement result based on its validity parameter for the corresponding performance parameters on the other. In an extreme case, this can mean that, for example, both a specificity and a sensitivity of 100% can be achieved for a virus test.

For example, if the biosensor is customized for a specific application scenario and a specificity of 98% and a validity of 59% are required with a threshold of 0.60, this threshold must be used to evaluate the measurement for this individual biosensor in order to achieve the required performance parameters.

In other words, if the measurement carried out results in a validity parameter that is below 0.95 when evaluating the measurement curve in the artificial intelligence 820 in step S5, the corresponding measurement is rejected as invalid. It is therefore always ensured that the parameters specified for the given application scenario are strictly adhered to by the measurement, or the measurement is rejected as invalid.

Irregularities in a measurement can lead to an unknown or irregular course of the measurement curve. By training the second artificial intelligence 820, a precise measurement result can still be determined for weak irregularities. A weak irregularity occurs, for example, if the measurement is aborted shortly before completion, or if weak disturbances, such as small temperature fluctuations, occur during the measurement. By repeatedly and adapted training the second artificial intelligence 820, such weak irregularities in the measurement can be compensated. Strong Irregularities can arise, for example, from a significantly premature termination of the measurement, whereby the termination causes the biosensor 1 to be separated from the specific analyte 90 significantly before the end of the measurement. With such an unknown or irregular course of the measurement curve, the second artificial intelligence 820 will also arrive at a result, but the validity parameter associated with the result will be very low. In other words, an irregular measurement will always result in the evaluation of the validity parameter in relation to the threshold in step S6 leading to the respective measurement being rejected as invalid. Consequently, repeated and adapted training of the second artificial intelligence 820 can reduce the number of measurements rejected as invalid.

Since the individualization of the respective biosensor is merely advantageous but not essential for the method shown in Figure 8 for determining the measurement result, the embodiment can also be viewed as starting with the method 100 for individualizing a biosensor 1. However, the individualization of the biosensor is not absolutely necessary.

Figure 9 uses a flow chart to show an embodiment for running through no or at least one further manufacturing step S7 of a biosensor 1 on the basis of a method 100 for individualizing the biosensor. As already described in Figure 4, successful individualization leads to a determination of an application scenario of the biosensor 1. In this case, both the application scenario and the performance parameters assigned to the application scenario with the respective threshold values can be stored in the biosensor 1. On the basis of the stored data, comprising the application scenario, the assigned performance parameters with the respective threshold values, biosensors, for example in stock as inventory, can run through at least one further manufacturing step S7. Manufacturing steps can include, for example, functionalization. During functionalization, a biosensor is adapted accordingly based on the stored data. The functionalization can include, among other things, applying a receptor layer 24 and/or reference layer 34. The production of a biosensor can also be stopped based on its individualization, so that the biosensor does not undergo any further production steps. The production steps can take place during production, during repair and/or during inspection of the biosensor. For example, biosensors can be flexibly adapted for their respective application scenario even after longer storage periods. Biosensors can also possibly undergo a process 100 for individualizing the biosensor at any time and then be reworked or their production can be stopped. Where applicable, all individual features shown in the exemplary embodiments can be combined with one another and/or exchanged without leaving the scope of the invention.

100 procedures

51 Determining at least one quality parameter

52 Determining at least one performance parameter and corresponding threshold

53 Customizing a Biosensor

54 Carrying out a measurement

55 Classifying the measurement

56 Evaluation of a measurement

57 Passing through no or at least one further production step

1 biosensor

2 test cantilevers

3 reference cantilevers

4 Electronics

9 sample

90 analytes

200 passive test transducers

220 active test transducers

300 passive reference transducers

320 active reference transducers

400, 401, 402, 403 electrodes

20 base

22 Deformable part

24 receptor layer

30 base

32 Deformable part

34 reference layer

44 bridge transverse voltage detector

440 AD converters

50 multimeters

52 measuring probes

60 camera system

62 computers 70 performance parameter list

72 shades of gray

80 neural networks

82 Neuron 84 First layer

85 Second Shift

86 Last Shift

88 training data

800 Measurement curve 820 Second artificial intelligence

822 measurement result

824 validity parameters

826 quality parameters

V0 excitation voltage VB bridge transverse voltage

Claims

claims 1. A method (100) for individualizing a biosensor (1) for measuring a specific analyte, comprising: Determining at least one quality parameter of the biosensor (1), Determining at least one performance parameter and an associated threshold value of the biosensor (1) based on the at least one quality parameter using a trained first artificial intelligence (S2), and Individualizing the biosensor to measure a specific analyte by evaluating at least one of the performance parameters and the associated threshold for an individual application scenario (S3). 2. Method according to claim 1, characterized in that the individualization of the biosensor (1) for measuring a specific analyte takes place using a trained first artificial intelligence taking into account a specific application scenario. 3. Method according to claim 1 or 2, characterized in that the quality parameter is determined on the basis of a property of the biosensor (1), wherein at least one electrical resistance is determined as a property, preferably all electrical resistances of a resistance bridge of the biosensor. 4. Method according to one of the preceding claims, characterized in that the biosensor (1) comprises at least one resistance bridge and at least one of the quality parameters is a self-similarity of the electrical resistances of the resistance bridge. 5. Method according to one of the preceding claims, characterized in that at least one of the performance parameters of the biosensor comprises a sensitivity and/or a specificity and/or a yield. 6. Method according to one of the preceding claims, characterized in that the trained first artificial intelligence determines at least one of the performance parameters and a Threshold value is determined on the basis of a feature vector, wherein the feature vector is obtained from a data preparation of the at least one quality parameter. 7. Method according to one of the preceding claims, characterized in that the trained first artificial intelligence is trained on the basis of training data obtained by the following steps: Determining a time-resolved measurement curve (800) of at least one specific analyte, as well as at least one sample without the specific analyte, each with a plurality of biosensors; Determining an individual measurement result (822) and a validity parameter (824) for each biosensor with a second artificial intelligence (820) from the time-resolved measurement curve (800) of the respective biosensor; Determining at least one quality parameter (826) for each biosensor; Statistically evaluating the individual measurement results (822) on the basis of the assigned validity parameters (824) using threshold values, the assigned quality parameters (826) and the previously known result for each biosensor; Determining performance parameters from statistical evaluation; Using the quality parameters, performance parameters and thresholds as training data. 8. Method according to claim 7, characterized in that the time-resolved measurement curve (800) is recorded at at least one defined concentration of the analyte and/or that the time-resolved measurement curve (800) is additionally recorded at a defined disturbance and/or that the time-resolved measurement curve (800) is recorded at at least one defined measurement routine. 9. Method according to one of the preceding claims, characterized in that the trained first artificial intelligence is trained on the basis of training data which include, as quality parameters, the self-similarity of the electrical resistances of a resistance bridge of each biosensor. 10. Method according to one of the preceding claims, characterized in that the trained first artificial intelligence is trained on the basis of training data which include as performance parameters the sensitivity and/or the specificity and/or the yield of each biosensor. 11. Method according to one of the preceding claims, characterized in that the biosensor, on the basis of its individualization, undergoes no or at least one further manufacturing step (S7), wherein the none or at least one further manufacturing step preferably comprises a storage of quality parameters and/or performance parameters and/or threshold values on the biosensor or a database and/or the at least one further manufacturing step comprises a functionalization of the biosensor, wherein the functionalization of the biosensor preferably comprises an application of a receptor layer and/or reference layer. 12. A method for generating training data for training a trained first artificial intelligence for use in the individualization of biosensors, comprising the following steps: Determining a time-resolved measurement curve (800) of at least one specific analyte, as well as at least one sample without the specific analyte, each with a plurality of biosensors; Determining an individual measurement result (822) and a validity parameter (824) for each biosensor with a second artificial intelligence (820) from the time-resolved measurement curve (800) of the respective biosensor; Determining at least one quality parameter (826) for each biosensor; Statistically evaluating the individual measurement results (822) on the basis of the assigned validity parameters (824) using threshold values, the assigned quality parameters (826) and the previously known result for each biosensor; Determining performance parameters from statistical evaluation; Using the quality parameters, performance parameters and thresholds as training data. 13. The method according to claim 12, characterized in that the time-resolved measurement curve (800) is recorded at at least one defined concentration of the analyte and/or that the time-resolved measurement curve (800) is additionally recorded at a defined disturbance and/or that the time-resolved measurement curve (800) is recorded at at least one defined measurement routine. 14. Method according to claim 12 or 13, characterized in that the quality parameter determined for each biosensor comprises the self-similarity of the resistances in a resistance bridge of each biosensor. 15. Biosensor for measuring a specific analyte, comprising at least one resistance bridge, wherein the biosensor is individualized for a specific application scenario for measuring the specific analyte based on a quality parameter determined from the resistance bridge. 16. Biosensor according to claim 15, characterized in that one of the quality parameters is the self-similarity of the resistances of the resistance bridge.