WO2024124497A1 - Machine-learning-based method for recognizing state of nanopore sequencing signal, and training method and apparatus for machine learning model - Google Patents

Abstract

A machine-learning-based method for recognizing the state of a nanopore sequencing signal. The method comprises: acquiring nanopore sequencing signal data X₁, X₂, …, X_N, and performing feature extraction on the nanopore sequencing signal data to generate a feature value vector; and by using the generated feature value vector and a trained machine learning model, recognizing the state of the nanopore sequencing signal data. The feature extraction comprises: dividing the nanopore sequencing signal data into G groups according to value magnitude, wherein G is a natural number greater than 1 and less than N, and a feature value comprises at least one of a quantity proportion, a standard deviation and a variation coefficient; calculating the quantity proportion of each of the G groups, so as to generate G quantity proportion feature values; calculating the standard deviation of data in each of the G groups, so as to generate G standard deviation feature values; calculating the variation coefficient of the data in each of the G groups, so as to generate G variation coefficient feature values; and generating a feature value vector on the basis of the plurality of generated feature values.

Description

Method for identifying nanopore sequencing signal state based on machine learning, method and device for training machine learning model Technical Field

The present invention relates to the field of bioinformatics, and more specifically to a method for identifying a nanopore sequencing signal state based on machine learning, a method for training a machine learning model, a device, an electronic device and a storage medium.

Background technique

Nanopore sequencing signals refer to the changes in the perforation current recorded when a DNA molecule passes through a nanopore. This change is mainly caused by the different currents of different bases in the nanopore. When analyzing nanopore sequencing signals, an important step is to perform state detection on the nanopore sequencing signals. The core task of state detection is to identify which state the nanopore sequencing signal is in, so as to provide a basis for the next analysis strategy.

Nanopore sequencing signals are one-dimensional time series signals. There are two existing technical solutions for detecting the state of such time series signals. One is the traditional time series data analysis method, which converts the time series data from the time domain to the frequency domain, and performs similarity calculation and threshold filtering on the frequency domain signal by manually extracting special sequence features to identify the state of the signal. The other is a deep learning end-to-end method, which does not require manual feature extraction, but directly trains a deep learning model based on the original data of the time series signal, and then uses this deep learning model to identify the signal state.

However, in some cases, traditional time series data analysis methods can only identify some target signals with obvious features that are easy to identify. In other cases, the robustness is very poor. For different types of target signals, it is necessary to design artificial feature extraction separately, which is inefficient. On the other hand, although the deep learning end-to-end method has good robustness, it has high computational complexity and needs to be run efficiently on the GPU. It will run very slowly on the CPU, and it takes a long time to train a deep learning model. Therefore, there is a need for a method, device and storage medium that can identify nanopore sequencing signals with lower complexity while improving robustness.

Summary of the invention

In order to solve the technical problems existing in the existing problems, in a first aspect of the present disclosure, a method for identifying a nanopore sequencing signal state based on machine learning is provided, comprising the following steps:

Acquire nanopore sequencing signal data X ₁ , X ₂ , . . . , X _N , and perform feature extraction on the nanopore sequencing signal data to generate a feature value vector, wherein N is a natural number greater than 1; and

Using the generated feature value vector and the trained machine learning model to identify the state of the acquired nanopore sequencing signal data,

The feature extraction comprises the following steps:

Dividing the nanopore sequencing signal data into G groups according to the size of the value, where G is a natural number greater than 1 and less than N; wherein the characteristic value includes at least one of the quantity proportion, standard deviation and coefficient of variation;

Calculate the quantity proportion of each group in the G groups respectively, and generate G quantity proportion feature values V ₁₁ , V ₁₂ , . . . , V _1G , wherein the quantity proportion of each group is the ratio of the quantity of data in the group to the total quantity of data N;

Calculating the standard deviation of the data of each group in the G groups respectively, and generating G standard deviation eigenvalues V ₂₁ , V ₂₂ , . . . , V _2G ;

Calculating the coefficient of variation of the data of each of the G groups respectively, generating G coefficient of variation characteristic values V ₃₁ , V ₃₂ , . . . , V _3G , wherein the coefficient of variation of the data of each group is the ratio of the standard deviation of the data of each group to the mean value of the data of each group; and

According to an embodiment of the present invention, before dividing the nanopore sequencing signal data into G groups according to the size of the values, the feature extraction further includes a standardization step, and the division is performed on the nanopore sequencing signal data after being standardized by the standardization step.

The standardization steps include:

For each data in the nanopore sequencing signal data, the difference between the value of the data and the minimum value of the signal data is divided by the difference between the maximum value of the signal data and the minimum value of the signal data, and the obtained value is used as the standardized value of the data, wherein the minimum value of the signal data is the minimum value in the acquired nanopore sequencing signal data, and the maximum value of the signal data is the maximum value in the acquired nanopore sequencing signal data.

According to an embodiment of the present invention, the division includes:

The normalized nanopore sequencing signal data is divided into 10 groups from 0 to 1 at intervals of 0.1, thereby dividing the signal data into 10 groups falling into the intervals [0, 0.1), [0.1, 0.2), ..., [0.9, 11] respectively.

According to an embodiment of the present invention, the feature extraction further comprises the following steps:

The maximum value, minimum value, average value, quotient of the maximum value and the minimum value, and quotient of the maximum value and the average value of the G standard deviation eigenvalues V ₂₁ , V ₂₂ , ..., V _2G are calculated to generate five eigenvalues V ₄₁ , V ₄₂ , ..., V ₄₅ .

According to an embodiment of the present invention, the feature extraction further comprises the following steps:

The maximum value, minimum value, average value, quotient of the maximum value and the minimum value, and quotient of the maximum value and the average value of the G variation coefficient eigenvalues V ₃₁ , V ₃₂ , ..., V _3G are calculated to generate five eigenvalues V ₅₁ , V ₅₂ , ..., V ₅₅ .

According to an embodiment of the present invention, the machine learning model includes a decision tree model, a Bayesian model or a neural network model.

In a second aspect of the present disclosure, a method for training a machine learning model for identifying nanopore sequencing signal states is provided, comprising the following steps:

Acquiring training data for the machine learning model, the acquiring comprising: collecting nanopore sequencing signal data, performing state recognition on the collected nanopore sequencing signal data, intercepting nanopore sequencing signal data _X1 , _X2 , ..., XN corresponding to the recognized state, performing feature extraction on the intercepted nanopore sequencing signal data _X1 , _X2 , ..., _XN to generate a eigenvalue vector, wherein N is a natural number greater than 1, and the training data comprises a result of the _state recognition and the eigenvalue vector; and

Using the acquired training data to train the machine learning model, thereby acquiring a trained machine learning model,

The feature extraction comprises the following steps:

Dividing the nanopore sequencing signal data into G groups according to the size of the value, wherein G is a natural number greater than 1 and less than N; wherein the characteristic value includes at least one of the quantity proportion, standard deviation and coefficient of variation;

Calculate the quantity proportion of each group in the G groups respectively, and generate G quantity proportion feature values V ₁₁ , V ₁₂ , . . . , V _1G , wherein the quantity proportion of each group is the ratio of the quantity of data in the group to the total quantity of data N;

Calculating the standard deviation of the data of each group in the G groups respectively, and generating G standard deviation eigenvalues V ₂₁ , V ₂₂ , . . . , V _2G ;

Calculating the coefficient of variation of the data of each of the G groups respectively, generating G coefficient of variation characteristic values V ₃₁ , V ₃₂ , . . . , V _3G , wherein the coefficient of variation of the data of each group is the ratio of the standard deviation of the data of each group to the mean value of the data of each group; and

The eigenvalue vector is generated based on the generated plurality of eigenvalues.

In a third aspect of the present disclosure, a device for identifying a nanopore sequencing signal state based on machine learning is provided, comprising the following modules:

an acquisition module, configured to acquire nanopore sequencing signal data X ₁ , X ₂ , . . . , X _N , and perform feature extraction on the nanopore sequencing signal data to generate a feature value vector, wherein N is a natural number greater than 1; and

an identification module, for identifying the state of the acquired nanopore sequencing signal data using the generated feature value vector and the trained machine learning model,

Wherein, the acquisition module further includes the following units:

A grouping unit, for dividing the nanopore sequencing signal data into G groups according to the size of the value, wherein G is a natural number greater than 1 and less than N; wherein the characteristic value includes at least one of the quantity proportion, standard deviation and coefficient of variation;

a first eigenvalue generating unit, configured to respectively calculate the quantity proportion of each group in the G groups, and generate G quantity proportion eigenvalues V ₁₁ , V ₁₂ , . . . , V _1G , wherein the quantity proportion of each group is the ratio of the quantity of data in the group to the total quantity of data N;

A second eigenvalue generating unit, configured to respectively calculate the standard deviation of the data of each of the G groups, and generate G standard deviation eigenvalues V ₂₁ , V ₂₂ , . . . , V _2G ;

a third eigenvalue generating unit, configured to respectively calculate a coefficient of variation of the data of each of the G groups, and generate G coefficient of variation eigenvalues V ₃₁ , V ₃₂ , . . . , V _3G , wherein the coefficient of variation of the data of each group is a ratio of the standard deviation of the data of each group to an average value of the data of each group; and

The eigenvalue vector generating unit is used to generate the eigenvalue vector based on the generated multiple eigenvalues.

In a fourth aspect of the present disclosure, there is provided a training device for a machine learning model for identifying a nanopore sequencing signal state, comprising the following modules:

an acquisition module, configured to acquire training data for the machine learning model, wherein the acquisition module acquires nanopore sequencing signal data, performs state recognition on the acquired nanopore sequencing signal data, intercepts nanopore sequencing signal data _X1 , _X2 , ..., _XN corresponding to the recognized state, performs feature extraction on the intercepted nanopore sequencing signal data _X1 , _X2 , ..., _XN to generate a eigenvalue vector, wherein N is a natural number greater than 1, and the training data includes the state recognition result and the eigenvalue vector; and

a training module, configured to train the machine learning model using the acquired training data, thereby acquiring a trained machine learning model,

Wherein, the acquisition module further includes the following units:

A grouping unit, for dividing the nanopore sequencing signal data into G groups according to the size of the value, wherein G is a natural number greater than 1 and less than N; wherein the characteristic value includes at least one of the quantity proportion, standard deviation and coefficient of variation;

a first eigenvalue generating unit, configured to respectively calculate the quantity proportion of each group in the G groups, and generate G quantity proportion eigenvalues V ₁₁ , V ₁₂ , . . . , V _1G , wherein the quantity proportion of each group is the ratio of the quantity of data in the group to the total quantity of data N;

A second eigenvalue generating unit, configured to respectively calculate the standard deviation of the data of each of the G groups, and generate G standard deviation eigenvalues V ₂₁ , V ₂₂ , . . . , V _2G ;

a third eigenvalue generating unit, configured to calculate the coefficient of variation of the data of each of the G groups, respectively, and generate G coefficient of variation eigenvalues V ₃₁ , V ₃₂ , . . . , V _3G , wherein the coefficient of variation of the data of each group is a ratio of the standard deviation of the data of each group to the mean value of the data of each group; and

The eigenvalue vector generating unit is used to generate the eigenvalue vector based on the generated multiple eigenvalues.

In a fourth aspect of the present disclosure, an electronic device is provided, comprising:

at least one processor; and

a memory communicatively connected to the at least one processor; wherein,

The memory stores instructions that can be executed by the at least one processor. The instructions are executed by the at least one processor to enable the at least one processor to perform any one of the methods described above.

In a fifth aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to cause the computer to execute any one of the methods described above.

Based on various aspects provided by the present disclosure, a method for identifying a nanopore sequencing signal state based on machine learning, a training method, an apparatus, an electronic device, and a non-transitory computer-readable storage medium for a machine learning model for identifying a nanopore sequencing signal state are provided. These aspects can identify nanopore sequencing signals with lower complexity while improving robustness.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a flowchart of a method for training a machine learning model for identifying nanopore sequencing signal states according to an embodiment of the present disclosure.

FIG. 2 is a schematic diagram of an exemplary image generated based on collected nanopore sequencing signal data.

FIG. 3 is a schematic diagram of three pictures obtained by classifying the exemplary picture shown in FIG. 2 through state recognition.

FIG4 is a flow chart of a method for identifying a nanopore sequencing signal state based on machine learning according to an embodiment of the present disclosure.

FIG5 shows an example of different states of a nanopore sequencing signal.

FIG. 6 shows another example of different states of a nanopore sequencing signal.

FIG. 7 shows yet another example of different states of a nanopore sequencing signal.

FIG8 is a schematic diagram of a device for identifying nanopore sequencing signal states based on machine learning according to an embodiment of the present disclosure.

FIG9 is a schematic diagram of a training device for a machine learning model for identifying nanopore sequencing signal states according to an embodiment of the present disclosure.

FIG10 shows an example of a decision tree model.

Detailed ways

The specific embodiments of the present invention will be described in detail below. It should be noted that the embodiments described herein are only for illustration and are not intended to limit the present invention. In the following description, a large number of specific details are set forth in order to provide a thorough understanding of the present invention. However, it is obvious to those of ordinary skill in the art that these specific details do not have to be used to implement the present invention. In other examples, in order to avoid confusing the present invention, known circuits, materials or methods are not specifically described.

Throughout the specification, references to "one embodiment," "an embodiment," "an example," or "an example" mean that a particular feature, structure, or characteristic described in conjunction with the embodiment or example is included in at least one embodiment of the present invention. Thus, the phrases "in one embodiment," "in an embodiment," "an example," or "an example" appearing in various places throughout the specification do not necessarily all refer to the same embodiment or example. Furthermore, particular features, structures, or characteristics may be combined in one or more embodiments or examples in any appropriate combinations and/or subcombinations.

It should be understood that when an element is said to be "coupled to" or "connected to" another element, it can be directly coupled or connected to the other element or there may be intermediate elements. On the contrary, when an element is said to be "directly coupled to" or "directly connected to" another element, there are no intermediate elements.

Furthermore, as used herein, the term "and/or" includes any and all combinations of one or more of the associated listed items.

It will be understood that a noun corresponding to a term in the singular form may include one or more things unless the relevant context clearly indicates otherwise. As used herein, each of phrases such as "A or B," "at least one of A and B," "at least one of A or B," "A, B or C," "at least one of A, B and C," and "at least one of A, B or C" may include all possible combinations of items listed together with a corresponding one of the plurality of phrases. As used herein, terms such as "1st" and "2nd" or "first" and "second" may be used to simply distinguish a corresponding component from another component and do not limit the components in other respects (e.g., importance or order).

As used herein, the term "module" may include a unit implemented in hardware, software, or firmware, and may be used interchangeably with other terms (e.g., "logic," "logic block," "portion," or "circuit"). A module may be a single integrated component adapted to perform one or more functions or a minimum unit or portion of the single integrated component. For example, according to an embodiment, a module may be implemented in the form of an application specific integrated circuit (ASIC).

It should be understood that the various embodiments of the present disclosure and the terms used therein are not intended to limit the technical features set forth herein to specific embodiments, but rather include various changes, equivalent forms or alternative forms for the corresponding embodiments. Unless otherwise expressly defined herein, all terms will be given their broadest possible interpretation, including the meanings implied in the specification and the meanings understood by those skilled in the art and/or defined in dictionaries, papers, etc.

In addition, it should be understood by those skilled in the art that the drawings provided herein are for illustrative purposes and are not necessarily drawn to scale. For the description of the drawings, similar reference numerals may be used to refer to similar or related elements. The present disclosure will be described exemplarily with reference to the drawings below.

In order to solve the problems of poor robustness of traditional time series data analysis methods and high computational complexity and cost of deep learning methods, the present disclosure proposes a new feature extraction method, and combines it with machine learning to train a machine learning model, and uses the machine learning model obtained in this way to identify the state of nanopore sequencing signals. The following is a detailed description of the method for identifying the state of nanopore sequencing signals based on machine learning, the training method of the machine learning model for identifying the state of nanopore sequencing signals, the device, the electronic device, and the non-transitory computer-readable storage medium according to the embodiments of the present invention with reference to the accompanying drawings.

Fig. 1 is a flow chart of a method for training a machine learning model for identifying the state of a nanopore sequencing signal according to an embodiment of the present disclosure. Referring to Fig. 1 , a method 100 for training a machine learning model for identifying the state of a nanopore sequencing signal according to an embodiment of the present disclosure is described below.

As shown in Figure 1, the training method 100 for a machine learning model for identifying the state of a nanopore sequencing signal according to an embodiment of the present disclosure includes steps S110 and S210. In step S110, training data for the machine learning model is obtained. In step S120, the machine learning model is trained using the obtained training data, thereby obtaining a trained machine learning model for identifying the state of a nanopore sequencing signal.

Specifically, step S110 further includes step S1101 and step S1102.

Wherein, in step S1101, nanopore sequencing signal data is collected.

For example, nanopore sequencing signal data can be collected by a nanopore sequencing device. When collecting signals, for example, a single-channel nanopore sequencing device and a computer can be used. The target sequencing signal data is collected by the nanopore sequencing device, and the collected sequencing signal data and other information are saved as an h5 file structure and stored in real time in a storage device of the computer.

The nanopore sequencing signal includes a state signal, which refers to a signal segment from the time when the pore current value decreases until it rises back to the pore current value, and the pore current value is known before the test data is collected.

The nanopore sequencing signal data may also be existing signal data collected in advance and stored in a computer system, rather than temporarily collected signal data, which is not specifically limited in the present disclosure.

In step S1102, training data is generated based on the collected nanopore sequencing signal data.

When making training data, first generate a picture based on the collected nanopore sequencing signal data. For example, with the current value as the vertical axis and the data collection sequence number (position in the data sequence) as the horizontal axis, the exemplary picture shown in Figure 2 can be generated based on the collected nanopore sequencing signal data. The purpose of generating these pictures is to manually classify the state signals, and the pictures are not used as inputs to the model. In the exemplary picture shown in Figure 2, the current signal at position 100 in the data sequence drops from the initial open hole current value and rises at position 500 (as shown by the reference numeral "1" in Figure 2), then drops from the open hole current value again at position 600 and rises at position 2000 (as shown by the reference numeral "2" in Figure 2), and drops from the open hole current value for the third time at position 3000 and rises at position 3500 (as shown by the reference numeral "3" in Figure 2). It can be seen that the exemplary picture shown in Figure 2 contains a total of 3 state signals.

Next, the generated images are manually classified. For example, by manually classifying the image shown in Figure 2, three categories can be obtained, named "0", "1" and "2" respectively. A single image is generated for each category, as shown in Figure 3. The starting position and ending position of the state signal of each category in the data sequence are recorded, and the image of each state signal is named "artificial classification name starting position in data sequence ending position in data sequence.png". In this way, according to the name of the image, the artificial classification name of the state signal image and the position of the state signal in the data sequence can be known, so as to intercept the corresponding data in the data sequence.

Next, according to the start position and end position of the data sequence of the picture name, the corresponding target data is intercepted in the data sequence, and features are extracted for the target data.

Feature extraction of target data may include the following steps.

First, the target data are grouped. Assuming that the target data is nanopore sequencing signal data X ₁ , X ₂ , ..., X _N , where N is a natural number greater than 1, these data are divided into G groups according to the value, where G is a natural number greater than 1 and less than N. The characteristic value includes at least one of the quantity proportion, standard deviation and coefficient of variation.

Next, the quantity proportion of each of the G groups is calculated respectively to generate G quantity proportion feature values V ₁₁ , V ₁₂ , . . . , V _1G , where the quantity proportion of each group is the ratio of the quantity of data in the group to the total quantity N of data.

Next, the standard deviation of the data of each of the G groups is calculated to generate G standard deviation eigenvalues V ₂₁ , V ₂₂ , . . . , V _2G .

Next, the coefficient of variation of the data of each of the G groups is calculated to generate G coefficient of variation characteristic values V ₃₁ , V ₃₂ , . . . , V _3G , wherein the coefficient of variation of the data of each group is the ratio of the standard deviation of the data of each group to the mean value of the data of each group.

Next, an eigenvalue vector is generated based on the generated multiple eigenvalues. For example, an eigenvalue vector V (V ₁₁ , V ₁₂ , ..., V _1G , V ₂₁ , V ₂₂ , ..., V _2G , V ₃₁ , V ₃₂ , ..., V _3G ) may be constructed based on the generated multiple eigenvalues.

Furthermore, before grouping, the feature extraction method may further include a normalization step to normalize the acquired nanopore sequencing signal data to a value in the [0, 1] interval.

After calculating the coefficient of variation eigenvalue, the feature extraction method may further include: calculating the maximum value, minimum value, average value, quotient of the maximum value and the minimum value, and quotient of the maximum value and the average value among the G standard deviation eigenvalues V ₂₁ , V ₂₂ , … , V _2G , to generate five eigenvalues V ₄₁ , V ₄₂ , … , V ₄₅ ; and calculating the maximum value, minimum value, average value, quotient of the maximum value and the minimum value, and quotient of the maximum value and the average value among the G coefficient of variation eigenvalues V ₃₁ , V ₃₂ , … , V _3G , to generate five eigenvalues V ₅₁ , V ₅₂ , … , V ₅₅ .

Therefore, when generating eigenvalue quantities, the constructed eigenvalue vector V can further include more eigenvalues, for example, it can be V (V ₁₁ , V ₁₂ , ... , V _1G , V ₂₁ , V ₂₂ , ... , V _2G , V ₃₁ , V ₃₂ , ... , V _3G , V ₄₁ , V ₄₂ , ... , V ₄₅ , V ₅₁ , V ₅₂ , ... , V ₅₅ ).

A specific example of the above feature extraction method is described in detail below.

(a) Standardize the target data (i.e., the status signal data), that is, for each data in the status signal data, divide the difference between the value of the data and the minimum value of the status signal data by the difference between the maximum value of the status signal data and the minimum value of the status signal data, and use the obtained value as the standardized value of the data.

(b) Grouping the standardized state signal data, that is, grouping the state signal data obtained in step (a) into 10 groups from 0 to 1 with an interval of 0.1, thereby dividing the state signal data into 10 groups falling into the intervals [0, 0.1), [0.1, 0.2), ..., [0.9, 11] respectively.

(c) Calculate the quantity ratio of each of the 10 groups divided in step (b) to generate 10 feature values about quantity ratio. The quantity ratio of each group refers to the ratio of the quantity of data in the group to the total quantity of data.

则这n个数据的标准差S为： (d) Calculate the standard deviation of each of the 10 groups obtained in step (b) and generate 10 eigenvalues related to the standard deviation. The standard deviation is calculated as follows. Suppose there are n data x ₁ , X ₂ , …, x _n , and their average is

Then the standard deviation S of these n data is:

(e) Calculate the coefficient of variation of each of the 10 groups divided in step (b) to generate 10 characteristic values for the coefficient of variation. The coefficient of variation is the ratio of the standard deviation to the mean, that is, coefficient of variation = standard deviation / mean.

(f) Calculate the maximum value, minimum value, average value, quotient of the maximum value and the minimum value, and quotient of the maximum value and the average value of the 10 eigenvalues generated in step (d) (i.e., eigenvalues with respect to the standard deviation) to generate 5 eigenvalues.

(g) Calculate the maximum value, minimum value, average value, quotient of the maximum value and the minimum value, and quotient of the maximum value and the average value of the 10 eigenvalues (i.e., eigenvalues regarding the coefficient of variation) generated in step (e) to generate 5 eigenvalues.

(h) A total of 40 eigenvalues are obtained in steps (c), (d), (e), (f), and (g), and these 40 eigenvalues are used to form a final eigenvalue vector as training data. For example, these 40 eigenvalues can be used as components of the eigenvalue vector in sequence to form a one-dimensional eigenvalue vector with 40 components. It should be noted that the method for forming the eigenvalue vector is not limited to this, and other methods commonly used by those skilled in the art can also be used.

Finally, each generated eigenvalue vector and the corresponding manual classification result are combined into a training data. For example, for the picture 0_100_500.png in Figure 3, the eigenvalue vector V1 is generated, and V1 and the corresponding manual classification result (the category named "0") constitute the first training data. Similarly, for the picture 1_600_2000.png in Figure 3, the eigenvalue vector V2 is generated, and V2 and the corresponding manual classification result (the category named "1") constitute the second training data. For the picture 2_3000_3500.png in Figure 3, the eigenvalue vector V3 is generated, and V3 and the corresponding manual classification result (the category named "2") constitute the third training data.

In step S120, the xgboost machine learning model is trained using the training data obtained in step S110, and the model file is saved.

During training, the training data and the corresponding manual classification results need to be divided into training sets and test sets. In the training set production stage, you can use random extraction and self-set ratios to produce a data set for machine learning training. Then, set the model training parameters according to the application scenario, such as learning rate and number of training iterations, and train and test the model. If the test results are not ideal, you can try to increase the number of data sets, modify the model training parameters, and adjust the training. Repeat this process until the model training results meet the indicator requirements.

Fig. 4 is a flow chart of a method for identifying a nanopore sequencing signal state based on machine learning according to an embodiment of the present disclosure. Referring to Fig. 4 , a method 400 for identifying a nanopore sequencing signal state based on machine learning according to an embodiment of the present disclosure is described below.

In step S410, nanopore sequencing signal data is acquired, and feature extraction is performed on the acquired nanopore sequencing signal data to generate a feature value vector. The acquired nanopore sequencing signal data is the nanopore sequencing signal data whose state is to be detected.

In order to obtain nanopore sequencing signal data, firstly, a single-channel nanopore sequencing device and a computer are used to collect sequencing signal data, and then signal segment data from the opening current value falling to the opening current value is intercepted in the sequencing signal data, and the signal segment data is used as the obtained nanopore sequencing signal data _X1 , _X2 , ..., _XN , where N is a natural number greater than 1, and the starting position and the ending position of the signal segment data in the entire nanopore sequencing signal data are recorded. The sequencing signal data may include one or more signal segment data, and the state identification is performed for each intercepted signal segment data.

The feature extraction method may include the following steps.

In the grouping step S4101, the acquired nanopore sequencing signal data _X1 , _X2 , ..., _XN are divided into G groups according to the size of the value, where G is a natural number greater than 1 and less than N. The characteristic value includes at least one of the quantity proportion, standard deviation and coefficient of variation.

In the quantity proportion eigenvalue calculation step S4102 , the quantity proportion of each of the G groups is calculated respectively to generate G quantity proportion eigenvalues V ₁₁ , V ₁₂ , . . . , V _1G , where the quantity proportion of each group is the ratio of the quantity of data in the group to the total quantity N of data.

In the standard deviation eigenvalue calculation step S4103 , the standard deviation of the data of each of the G groups is calculated respectively to generate G standard deviation eigenvalues V ₂₁ , V ₂₂ , . . . , V _2G .

In the variation coefficient characteristic value calculation step S4104, the variation coefficient of the data of each of the G groups is calculated respectively to generate G variation coefficient characteristic values V ₃₁ , V ₃₂ , ..., V _3G , where the variation coefficient of each group of data is the ratio of the standard deviation of each group of data to the mean value of each group of data.

In the feature value generating step S4105, a feature value vector is generated based on the generated multiple feature values. For example, a feature value vector V (V ₁₁ , V ₁₂ , ..., V _1G , V ₂₁ , V ₂₂ , ..., V _2G , V ₃₁ , V ₃₂ , ..., V _3G ) can be constructed based on the generated multiple feature values.

In addition, before the grouping step S4101, the feature extraction method may further include a normalization step to normalize the acquired nanopore sequencing signal data to a value in the interval [0, 1].

After the variation coefficient eigenvalue calculation step S4104, the feature extraction method may further include: calculating the maximum value, minimum value, average value, quotient of the maximum value and the minimum value, and quotient of the maximum value and the average value among the G standard deviation eigenvalues V ₂₁ , V ₂₂ , … , V _2G , and generating five eigenvalues V ₄₁ , V 42 , … , V ₄₅ ; and calculating the maximum value, minimum value, average value, quotient of the maximum value and the minimum value, and quotient of _{the maximum value and the average value among the G variation coefficient eigenvalues V 31 ,} V ₃₂ , … , V _3G , and generating five eigenvalues V ₅₁ , V ₅₂ , … , V ₅₅ .

Therefore, in the eigenvalue generating step S4105, the constructed eigenvalue vector V can further include more eigenvalues, for example, it can be V (V ₁₁ , V ₁₂ , ... , V _1G , V ₂₁ , V ₂₂ , ... , V _2G , V ₃₁ , V ₃₂ , ... , V _3G , V ₄₁ , V ₄₂ , ... , V ₄₅ , V ₅₁ , V ₅₂ , ... , V ₅₅ ).

A specific example of the feature extraction method is the same as steps (a) to (h) described above, and will not be repeated here.

In step S420, the generated eigenvalue vector and the trained machine learning model are used to identify the state of the acquired nanopore sequencing signal data. For example, the trained machine learning model can be a model trained by the training method of the machine learning model according to the embodiment of the present disclosure shown in FIG. 1.

Specifically, the feature value vector extracted in step S410 is input into the trained machine learning model to obtain a prediction result. For example, the prediction result can be one of the three values 0, 1, and 2. In addition, a picture of each state signal identified can be output according to the prediction result. For example, the picture can be named "prediction result name_starting position in nanopore sequencing signal data_ending position in nanopore sequencing signal data.png". The names and formats of the pictures described here are only examples, and the present disclosure is not limited thereto.

Figures 5 to 7 show examples of different states of nanopore sequencing signals. It can be seen that some states contain a large amount of data, and it takes a lot of time and effort to manually determine the states of all signals, and the accuracy of the determination results is not high. The machine learning model trained using the training method of the present disclosure can identify the states of a large number of signals with high accuracy, thereby improving efficiency and saving time.

FIG8 is a schematic diagram of a device for identifying a nanopore sequencing signal state based on machine learning according to an embodiment of the present disclosure. Referring to FIG8 , a device 800 for identifying a nanopore sequencing signal state based on machine learning according to an embodiment of the present disclosure is described below. The identification device 800 includes an acquisition module 810 and an identification module 820.

The acquisition module 810 is used to acquire nanopore sequencing signal data X ₁ , X ₂ , . . . , X _N , and perform feature extraction on the nanopore sequencing signal data to generate a feature value vector, wherein N is a natural number greater than 1.

The identification module 820 is used to identify the state of the acquired nanopore sequencing signal data using the generated eigenvalue vector and the trained machine learning model.

The acquisition module 810 further includes the following units: a grouping unit 8101, which is used to divide the nanopore sequencing signal data into G groups according to the size of the value, where G is a natural number greater than 1 and less than N; wherein the characteristic value includes the number proportion, standard deviation and variation coefficient. at least one of the numbers; a first eigenvalue generating unit 8102, used to respectively calculate the quantity proportion of each group in the G groups, and generate G quantity proportion eigenvalues V_11, V_12, ..., V_1G, wherein the quantity proportion of each group is the ratio of the quantity of data in the group to the total quantity of data N; a second eigenvalue generating unit 8103, used to respectively calculate the standard deviation of the data of each group in the G groups, and generate G standard deviation eigenvalues V_21, V_22, ..., V_2G; a third eigenvalue generating unit 8104, used to respectively calculate the coefficient of variation of the data of each group in the G groups, and generate G coefficient of variation eigenvalues V_31, V_32, ..., V_3G, wherein the coefficient of variation of the data of each group is the ratio of the standard deviation of the data of each group to the average value of the data of each group; and an eigenvalue vector generating unit 8105, used to generate the eigenvalue vector based on the generated multiple eigenvalues.

FIG9 is a schematic diagram of a training device for a machine learning model for identifying a nanopore sequencing signal state according to an embodiment of the present disclosure. Referring to FIG9 , a training device 900 for a machine learning model for identifying a nanopore sequencing signal state according to an embodiment of the present disclosure is described below. The training device 900 includes an acquisition module 910 and a training module 920.

The acquisition module 910 is used to acquire training data for a machine learning model, wherein the acquisition module 910 acquires nanopore sequencing signal data, performs state recognition on the acquired nanopore sequencing signal data, intercepts nanopore sequencing signal data _X1 , _X2 , ..., _XN corresponding to the recognized state, performs feature extraction on the intercepted nanopore sequencing signal data _X1 , _X2 , ..., _XN to generate a eigenvalue vector, wherein N is a natural number greater than 1, and the training data includes the state recognition result and the eigenvalue vector.

The training module 920 is used to train the machine learning model using the acquired training data, thereby obtaining a trained machine learning model.

The acquisition module 910 further includes the following units: a grouping unit 9101, used to divide the nanopore sequencing signal data into G groups according to the size of the value, wherein G is a natural number greater than 1 and less than N; wherein the eigenvalue includes at least one of the quantity proportion, standard deviation and coefficient of variation; a first eigenvalue generating unit 9102, used to respectively calculate the quantity proportion of each group in the G groups, and generate G quantity proportion eigenvalues V ₁₁ , V ₁₂ , ..., V _1G , wherein the quantity proportion of each group is the ratio of the quantity of data in the group to the total quantity of data N; a second eigenvalue generating unit 9103, used to respectively calculate the standard deviation of the data of each group in the G groups, and generate G standard deviation eigenvalues V ₂₁ , V ₂₂ , ..., V _2G ; a third eigenvalue generating unit 9104, used to respectively calculate the coefficient of variation of the data of each group in the G groups, and generate G coefficient of variation eigenvalues V ₃₁ , V ₃₂ , ..., V _3G , wherein the coefficient of variation of each group of data is the ratio of the standard deviation of each group of data to the mean value of each group of data; and an eigenvalue vector generating unit 9105, which is used to generate the eigenvalue vector based on the generated multiple eigenvalues.

The machine learning model in the present disclosure includes models such as decision tree models, Bayesian models or neural network models.

A specific example of the machine learning model in the present disclosure may be an open source xgboost model. For specific details, please refer to the open source code. The xgboost model is briefly described below.

Xgboost is an additive formula consisting of k base models:

是第i个样本x _i的预测值。 where _ft is the kth basis model,

is the predicted value of the ith sample _xi .

Each base model is a decision tree model. FIG10 shows an example of a decision tree model. As shown in FIG10 , whether people like computer games can be classified according to their gender and age. For example, when the age is greater than or equal to 15, the score is -1; when the age is less than 15, if the person is male, the score is +2, and if the person is female, the score is +0.1. Among them, the higher the score, the more people like computer games.

The feature extraction method described in the present disclosure is not limited to the above-mentioned specific embodiments, and those skilled in the art can make various improvements, deformations, deletions and substitutions based on the feature extraction method of the present disclosure. For example, the various eigenvalues included in the generated eigenvalue vector can be increased, decreased or replaced. As an example, one or more of the maximum value, minimum value, average value, quotient of the maximum value and the minimum value, and quotient of the maximum value and the average value of 10 quantity-proportion eigenvalues can be calculated, and the calculated values can be added to the generated eigenvalue vector.

The machine learning model described in the present disclosure is not limited to the above specific embodiments. For example, common models such as linear regression, support vector machine, random forest, clustering algorithm, hidden Markov model, etc. can also be used.

The method for identifying the state of a nanopore sequencing signal based on machine learning and the method for training a machine learning model described in the present disclosure are not limited to state identification of nanopore sequencing signals, but can also be extended to the field of classification and identification of other time domain and frequency domain signals.

Although multiple components are shown in the above block diagrams, those skilled in the art will appreciate that the embodiments of the present invention may be implemented without one or more components or by combining certain components.

Although the various steps are described above according to the order shown in the drawings, those skilled in the art should understand that the various steps may be performed in a different order, or the embodiments of the present invention may be implemented without one or more of the above steps.

It can be understood from the foregoing that the electronic components of one or more systems or devices may include, but are not limited to, at least one processing unit, a memory, and a communication bus or communication device that couples the various components including the memory to the processing unit. The system or device may include or may access various device-readable media. The system memory may include device-readable storage media in the form of volatile and/or non-volatile memory (e.g., read-only memory (ROM) and/or random access memory (RAM)). By way of example and not limitation, the system memory may also include an operating system, application programs, other program modules, and program data.

Embodiments may be implemented as systems, methods, or program products. Thus, embodiments may take the form of all-hardware embodiments or embodiments that include software (including firmware, resident software, microcode, etc.), which may be collectively referred to herein as "circuits," "modules," or "systems." In addition, embodiments may take the form of a program product embodied in at least one device-readable medium having device-readable program code embodied thereon.

Combinations of device-readable storage media may be used. In the context of this document, a device-readable storage medium ("storage medium") may be any tangible, non-signal medium that may contain or store a program consisting of program code configured to be used by or in conjunction with an instruction execution system, apparatus, or device. For purposes of this disclosure, a storage medium or device should be interpreted as being non-transitory, i.e., not including signal or propagation media.

The present disclosure is presented for purposes of illustration and description, but is not intended to be exhaustive or limiting. Many modifications and variations will be apparent to those of ordinary skill in the art. The embodiments are chosen and described in order to illustrate the principles and practical applications, and to enable those of ordinary skill in the art to understand the various embodiments of the present disclosure with various modifications suitable for the particular use contemplated.

Claims (16)

A method for identifying a nanopore sequencing signal state based on machine learning comprises the following steps: Acquire nanopore sequencing signal data X ₁ , X ₂ , . . . , X _N , and perform feature extraction on the nanopore sequencing signal data to generate a feature value vector, wherein N is a natural number greater than 1; and Using the generated feature value vector and the trained machine learning model to identify the state of the acquired nanopore sequencing signal data, The feature extraction comprises the following steps: Dividing the nanopore sequencing signal data into G groups according to the size of the value, where G is a natural number greater than 1 and less than N; wherein the characteristic value includes at least one of the quantity proportion, standard deviation and coefficient of variation; Calculate the quantity proportion of each group in the G groups respectively, and generate G quantity proportion feature values V ₁₁ , V ₁₂ , . . . , V _1G , wherein the quantity proportion of each group is the ratio of the quantity of data in the group to the total quantity of data N; Calculating the standard deviation of the data of each group in the G groups respectively, and generating G standard deviation eigenvalues V ₂₁ , V ₂₂ , . . . , V _2G ; Calculating the coefficient of variation of the data of each of the G groups respectively, generating G coefficient of variation characteristic values V _31, V ₃₂ , . . . , V _3G , wherein the coefficient of variation of the data of each group is the ratio of the standard deviation of the data of each group to the mean value of the data of each group; and The eigenvalue vector is generated based on the generated plurality of eigenvalues. The identification method according to claim 1, wherein, before dividing the nanopore sequencing signal data into G groups according to the size of the value, the feature extraction further includes a standardization step, and the division is performed on the nanopore sequencing signal data standardized by the standardization step, The standardization steps include: For each data in the nanopore sequencing signal data, the difference between the value of the data and the minimum value of the signal data is divided by the difference between the maximum value of the signal data and the minimum value of the signal data, and the obtained value is used as the standardized value of the data, wherein the minimum value of the signal data is the minimum value in the acquired nanopore sequencing signal data, and the maximum value of the signal data is the maximum value in the acquired nanopore sequencing signal data. The identification method according to claim 2, wherein the division comprises: The normalized nanopore sequencing signal data is divided into 10 groups from 0 to 1 at intervals of 0.1, thereby dividing the signal data into 10 groups falling into the intervals [0, 0.1), [0.1, 0.2), ..., [0.9, 1] respectively. The recognition method according to claim 1, wherein the feature extraction further comprises the following steps: The maximum value, minimum value, average value, quotient of the maximum value and the minimum value, and quotient of the maximum value and the average value of the G standard deviation eigenvalues V ₂₁ , V ₂₂ , ..., V _2G are calculated to generate five eigenvalues V ₄₁ , V ₄₂ , ..., V ₄₅ . The recognition method according to claim 1, wherein the feature extraction further comprises the following steps: The maximum value, minimum value, average value, quotient of the maximum value and the minimum value, and quotient of the maximum value and the average value of the G variation coefficient eigenvalues V ₃₁ , V ₃₂ , ..., V _3G are calculated to generate five eigenvalues V ₅₁ , V ₅₂ , ..., V ₅₅ . The recognition method according to claim 1, wherein the machine learning model comprises a decision tree model, a Bayesian model or a neural network model. A method for training a machine learning model for identifying nanopore sequencing signal states, comprising the following steps: Acquiring training data for the machine learning model, the acquiring comprising: collecting nanopore sequencing signal data, performing state recognition on the collected nanopore sequencing signal data, intercepting nanopore sequencing signal data _X1 , _X2 , ..., XN corresponding to the recognized state, performing feature extraction on the intercepted nanopore sequencing signal data _X1 , _X2 , ..., _XN to generate a eigenvalue vector, wherein N is a natural number greater than 1, and the training data comprises a result of the _state recognition and the eigenvalue vector; and Using the acquired training data to train the machine learning model, thereby acquiring a trained machine learning model, The feature extraction comprises the following steps: Dividing the nanopore sequencing signal data X ₁ , X ₂ , ..., X _N into G groups according to the size of the values, wherein G is a natural number greater than 1 and less than N; wherein the characteristic value includes at least one of the quantity proportion, the standard deviation and the coefficient of variation; Calculate the quantity proportion of each group in the G groups respectively, and generate G quantity proportion feature values V ₁₁ , V ₁₂ , . . . , V _1G , wherein the quantity proportion of each group is the ratio of the quantity of data in the group to the total quantity of data N; Calculating the standard deviation of the data of each group in the G groups respectively, and generating G standard deviation eigenvalues V ₂₁ , V ₂₂ , . . . , V _2G ; Calculating the coefficient of variation of the data of each of the G groups respectively, generating G coefficient of variation characteristic values V ₃₁ , V ₃₂ , . . . , V _3G , wherein the coefficient of variation of the data of each group is the ratio of the standard deviation of the data of each group to the mean value of the data of each group; and The eigenvalue vector is generated based on the generated plurality of eigenvalues. The training method according to claim 7, wherein, before dividing the nanopore sequencing signal data into G groups according to the size of the value, the feature extraction further includes a standardization step, and the division is performed on the nanopore sequencing signal data standardized by the standardization step, The standardization steps include: For each data in the nanopore sequencing signal data, the difference between the value of the data and the minimum value of the signal data is divided by the difference between the maximum value of the signal data and the minimum value of the signal data, and the obtained value is used as the standardized value of the data, wherein the minimum value of the signal data is the minimum value in the acquired nanopore sequencing signal data, and the maximum value of the signal data is the maximum value in the acquired nanopore sequencing signal data. The training method according to claim 8, wherein the dividing comprises: The normalized nanopore sequencing signal data is divided into 10 groups from 0 to 1 at intervals of 0.1, thereby dividing the signal data into 10 groups falling into the intervals [0, 0.1), [0.1, 0.2), ..., [0.9, 1] respectively. The training method according to claim 7, wherein the feature extraction further comprises the following steps: The maximum value, minimum value, average value, quotient of the maximum value and the minimum value, and quotient of the maximum value and the average value of the G standard deviation eigenvalues V ₂₁ , V ₂₂ , ..., V _2G are calculated to generate five eigenvalues V ₄₁ , V ₄₂ , ..., V ₄₅ . The training method according to claim 7, wherein the feature extraction further comprises the following steps: The maximum value, minimum value, average value, quotient of the maximum value and the minimum value, and quotient of the maximum value and the average value of the G variation coefficient eigenvalues V ₃₁ , V ₃₂ , ..., V _3G are calculated to generate five eigenvalues V ₅₁ , V ₅₂ , ..., V ₅₅ . The training method according to claim 7, wherein the machine learning model comprises a decision tree model, a Bayesian model or a neural network model. A device for identifying the state of a nanopore sequencing signal based on machine learning, comprising the following modules: an acquisition module, configured to acquire nanopore sequencing signal data X ₁ , X ₂ , . . . , X _N , and perform feature extraction on the nanopore sequencing signal data to generate a feature value vector, wherein N is a natural number greater than 1; and an identification module, for identifying the state of the acquired nanopore sequencing signal data using the generated feature value vector and the trained machine learning model, Wherein, the acquisition module further includes the following units: A grouping unit, configured to divide the nanopore sequencing signal data into G groups according to the size of the value, wherein G is a natural number greater than 1 and less than N; wherein the characteristic value includes at least one of a quantity proportion, a standard deviation, and a coefficient of variation; a first eigenvalue generating unit, configured to respectively calculate the quantity proportion of each group in the G groups, and generate G quantity proportion eigenvalues V ₁₁ , V ₁₂ , . . . , V _1G , wherein the quantity proportion of each group is the ratio of the quantity of data in the group to the total quantity of data N; A second eigenvalue generating unit, configured to respectively calculate the standard deviation of the data of each of the G groups, and generate G standard deviation eigenvalues V ₂₁ , V ₂₂ , . . . , V _2G ; a third eigenvalue generating unit, configured to calculate the coefficient of variation of the data of each of the G groups, respectively, and generate G coefficient of variation eigenvalues V ₃₁ , V ₃₂ , . . . , V _3G , wherein the coefficient of variation of the data of each group is a ratio of the standard deviation of the data of each group to the mean value of the data of each group; and The eigenvalue vector generating unit is used to generate the eigenvalue vector based on the generated multiple eigenvalues. A training device for a machine learning model for identifying nanopore sequencing signal states, comprising the following modules: an acquisition module, configured to acquire training data for the machine learning model, wherein the acquisition module acquires nanopore sequencing signal data, performs state recognition on the acquired nanopore sequencing signal data, intercepts nanopore sequencing signal data _X1 , _X2 , ..., _XN corresponding to the recognized state, performs feature extraction on the intercepted nanopore sequencing signal data _X1 , _X2 , ..., _XN to generate a eigenvalue vector, wherein N is a natural number greater than 1, and the training data includes the state recognition result and the eigenvalue vector; and a training module, configured to train the machine learning model using the acquired training data, thereby acquiring a trained machine learning model, Wherein, the acquisition module further includes the following units: A grouping unit, for dividing the nanopore sequencing signal data X ₁ , X ₂ , ..., X _N into G groups according to the size of the values, wherein G is a natural number greater than 1 and less than N; wherein the characteristic value includes at least one of the quantity proportion, standard deviation and coefficient of variation; a first eigenvalue generating unit, configured to respectively calculate the quantity proportion of each group in the G groups, and generate G quantity proportion eigenvalues V ₁₁ , V ₁₂ , . . . , V _1G , wherein the quantity proportion of each group is the ratio of the quantity of data in the group to the total quantity of data N; A second eigenvalue generating unit, configured to respectively calculate the standard deviation of the data of each of the G groups, and generate G standard deviation eigenvalues V ₂₁ , V ₂₂ , . . . , V _2G ; a third eigenvalue generating unit, configured to calculate the coefficient of variation of the data of each of the G groups, respectively, and generate G coefficient of variation eigenvalues V ₃₁ , V ₃₂ , . . . , V _3G , wherein the coefficient of variation of the data of each group is a ratio of the standard deviation of the data of each group to the mean value of the data of each group; and The eigenvalue vector generating unit is used to generate the eigenvalue vector based on the generated multiple eigenvalues. An electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 12. A non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause the computer to execute the method according to any one of claims 1-12.

Images