WO2022266928A1

WO2022266928A1 - Metabolic characteristic spectrum inference method and system, and computer device and storage medium

Info

Publication number: WO2022266928A1
Application number: PCT/CN2021/102060
Authority: WO
Inventors: 李伟忠; 邓永洁; 胡寓旻; 黄蓬
Original assignee: 中山大学
Priority date: 2021-06-24
Filing date: 2021-06-24
Publication date: 2022-12-29

Abstract

Disclosed in the present invention are a metabolic characteristic spectrum inference method. The method comprises: performing LC-MS technical processing on target sample data, so as to obtain LC-MS original data; performing dimension-reduction conversion processing on the LC-MS original data, so as to obtain a two-dimensional matrix, wherein the two-dimensional matrix retains the retention time, mass-to-charge ratio and ionic strength of the LC-MS original data; and inputting the two-dimensional matrix into a convolutional neural network model so as to infer a metabolic substance characteristic spectrum of the target sample data. Further disclosed in the present invention are a metabolic characteristic spectrum inference system, a computer device and a storage medium. By using the present invention, the problems in the existing metabonomic methods of it being difficult to process an error, a large number of original signals being lost, and the limitations of large-class differentiation can be solved.

Description

Metabolic profile deduction method, system, computer equipment and storage medium

technical field

The invention relates to the field of metabolomics data analysis, in particular to a method, system, computer equipment and storage medium for inferring metabolic profile.

Background technique

Metabolites in human serum include host metabolites, microbial-derived metabolites, and exogenous substances such as diet, which are closely related to the occurrence and development of various diseases. Current metabolomics methods are capable of quantitative determination, identification and analysis of metabolites in serum. Liquid Chromatograph-Mass Spectrometer (LC-MS) is a commonly used detection technology for metabolites. Different substances are separated by high-performance liquid chromatography, and mass spectrometry is used to analyze the quality of the substances separated in different phases. . At present, the identification of substances in non-targeted LC-MS raw data is mainly carried out through database comparison. substances for comparison. Among them, the Human Metabolome Database (The Human Metabolome Database, HMDB) contains 114,305 metabolite entries. But they are still few compared to the actual chemical space. More than 166 billion small organic molecules are listed in the Chemical Universe database GDB-17. Furthermore, there are several challenges in the processing of metabolomics data (i.e., sparse, noisy, heterogeneous, time-dependent, etc.). At this stage, deep learning techniques are rarely used in metabolomics data. The SteroidXtract tool applies deep learning technology and can directly use the original mass spectrum to classify steroid substances and non-steroid substances. However, LC-MS data is a kind of complex three-dimensional spatial data. The same sample contains multiple time-phase data (ie, different retention times), and each time-phase data has a mass spectrum. The SteroidXtract method and other metabolomics analysis methods require artificial de-redundancy processing of these large mass spectra. In addition, the biological processes involved in metabolites in serum are often not associated with a single type or substance, and these different substances are often distributed in different phases.

Traditional metabolomics methods first need to go through complex processes to remove noise and extract signal mass spectrum peaks, and then use statistical methods and rely on existing databases for correlation analysis and substance identification. First of all, in the process of data processing, problems such as data sparsity, noise, and batch effects have brought a lot of errors to mass spectrum peak-to-peak alignment, mass spectrum peak extraction, and subsequent statistical analysis. Secondly, the existing databases cannot cover a large number of metabolites in the real chemical world, and some unknown metabolites may also play an important role in the occurrence and development of diseases. Existing deep learning techniques use mass spectra as input data, which not only requires complicated de-redundancy processing, but also can only distinguish between steroids and non-steroids. Some isomers with different functions, etc. may have similar mass spectrometry performance, but are separated into different phases in liquid chromatography.

Contents of the invention

The technical problem to be solved by the present invention is to provide a method, system, computing device, and storage medium for inferring metabolic profiles, which can solve the difficulties in error handling, loss of a large number of original signals, and the limitations of classification in existing metabolomics methods. sex issue.

In order to solve the above technical problems, the present invention provides a method for inferring metabolic profiles, comprising: subjecting target sample data to LC-MS technology processing to obtain LC-MS raw data; performing dimensionality reduction transformation on the LC-MS raw data Processing to obtain a two-dimensional matrix that retains the retention time, mass-to-charge ratio, and ion intensity of the LC-MS raw data; inputting the two-dimensional matrix into a convolutional neural network model to infer the target sample Metabolite profiles of the data.

Preferably, the step of performing dimensionality reduction conversion processing on the LC-MS raw data to obtain a two-dimensional matrix includes: performing format conversion on the LC-MS raw data; setting the initial retention time, termination retention time, retention Time interval, retention time sampling interval, initial mass-to-charge ratio, termination mass-to-charge ratio, mass-to-charge ratio interval, and mass-to-charge ratio sampling interval, wherein the retention time interval ranges from the initial retention time to the termination retention time The range of the mass-to-charge ratio is the range between the initial mass-to-charge ratio and the end mass-to-charge ratio; in the retention time interval and the mass-to-charge ratio interval, the retention time sampling interval and mass-to-charge ratio sampling The interval is a sliding window, and the maximum ion intensity within the retention time interval and mass-to-charge ratio interval is sampled to obtain a two-dimensional matrix of ion intensity.

Preferably, the step of inputting the two-dimensional matrix into the convolutional neural network model to infer the metabolite profile of the sample includes: performing class activation thermodynamic calculations according to the convolutional neural network model to generate This class activation score s (t, r), wherein, t is the retention time, r is the mass-to-charge ratio; according to the network structure of the convolutional neural network model, the mapping function is extracted: t=map1(x), r= map2(y), wherein, t is the retention time, and r is the mass-to-charge ratio; according to the mapping function, the two-dimensional coordinates of the class activation thermodynamic map are mapped to retention time and mass-to-charge ratio; the class activation fraction is Filter processing to obtain retention characteristics; screen key metabolites according to the retention characteristics, and perform correlation calculations to infer the metabolic markers and metabolic network patterns of the target sample data, and then generate the metabolic profile of the target sample data .

Preferably, the step of filtering the class activation scores to obtain retained features comprises: filtering out molecular features whose class activation scores are less than a first preset threshold and whose ion intensity is less than a second preset threshold, to obtain Preserve features.

The present invention also provides a metabolic profile inference system, including: an LC-MS processing module, used to process target sample data with LC-MS technology to obtain LC-MS raw data; a dimensionality reduction conversion processing module, used to convert The LC-MS raw data is subjected to dimension reduction conversion processing to obtain a two-dimensional matrix, and the two-dimensional matrix retains the retention time, mass-to-charge ratio and ion intensity of the LC-MS raw data; the metabolic profile deduction module is used for The two-dimensional matrix is input into the convolutional neural network model to infer the metabolic substance characteristic spectrum of the target sample data.

Preferably, the dimensionality reduction conversion processing module includes: a format conversion unit for performing format conversion on the LC-MS raw data; a parameter setting unit for setting the initial retention time, termination retention time, retention time interval, Retention time sampling interval, initial mass-to-charge ratio, termination mass-to-charge ratio, mass-to-charge ratio interval, and mass-to-charge ratio sampling interval, wherein the retention time interval ranges from the initial retention time to the termination retention time, The mass-to-charge ratio interval is the range between the initial mass-to-charge ratio and the end mass-to-charge ratio; the dimensionality reduction sampling unit is used for sampling intervals of the retention time and The mass-to-charge ratio sampling interval is a sliding window, and the maximum ion intensity within the retention time interval and the mass-to-charge ratio interval is sampled to obtain a two-dimensional matrix of ion intensity.

Preferably, the metabolic profile inference module includes: a class activation score acquisition unit, which is used to calculate the class activation heat map according to the convolutional neural network model, and generate a class activation score s(t, r) for each sample, Wherein, t is the retention time, r is the mass-to-charge ratio; the extraction unit is used to extract the mapping function according to the network structure of the convolutional neural network model: t=map1(x), r=map2(y), wherein, T is the retention time, r is the mass-to-charge ratio; the mapping unit is used to map the two-dimensional coordinates of the class activation thermodynamic map to retention time and mass-to-charge ratio according to the mapping function; the filter unit is used to classify the class The activation score is filtered to obtain the retention characteristics; the calculation inference unit is used to screen key metabolites according to the retention characteristics, and perform correlation calculations to infer the metabolic markers and metabolic network patterns of the target sample data, and then generate The metabolic profile of the target sample data.

Preferably, the filtering unit is configured to filter out molecular features whose class activation scores are smaller than a first preset threshold and whose ion intensity is smaller than a second preset threshold, so as to obtain reserved features.

The present invention also provides a computer device, which includes a memory, a processor, and computer instructions stored in the memory and operable on the processor. The steps of the above method are realized when the processor executes the instructions.

The present invention also provides a storage medium, which stores computer instructions, and when the computer instructions are executed by a processor, the steps of the above method are realized.

The beneficial effect of implementing the present invention is:

In the present invention, the raw data of LC-MS is obtained by first processing the sample that needs to infer the metabolic profile through LC-MS technology, wherein the LC-MS technology is liquid chromatography mass spectrometry technology; and then the LC-MS technology is The MS raw data is subjected to dimensionality reduction conversion processing to obtain a two-dimensional matrix, and finally the two-dimensional matrix is input into the convolutional neural network model to infer the metabolic substance characteristic spectrum of the sample.

With the present invention, the two-dimensional conversion processing of LC-MS raw data can effectively reduce the data size, thereby contributing to subsequent calculations; compared with the existing methods that will cause a large number of signal loss when removing redundancy, the LC-MS The two-dimensional conversion processing of MS raw data can preserve the material signal to the greatest extent; the present invention extracts sample attribute-related features from the final convolutional neural network model, which can more effectively evaluate the joint correlation between multiple substances and sample classification, and Instead of comparing each substance one by one in isolation, it is possible to more accurately infer the metabolic profile of the sample.

Description of drawings

Fig. 1 is the flowchart of the metabolic profile deduction method provided by the present invention;

Fig. 2 is the flow chart of the method for dimension reduction conversion processing provided by the present invention;

Fig. 3 is the flow chart of the method for inferring the profile of metabolites provided by the present invention;

Fig. 4 is a schematic diagram of the method for inferring metabolic profiles provided by the present invention;

Fig. 5 is a functional block diagram of the metabolic profile inference system provided by the present invention;

Fig. 6 is a functional block diagram of the dimension reduction conversion processing module provided by the present invention;

Fig. 7 is a functional block diagram of the metabolic profile inference module provided by the present invention.

detailed description

In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings. It is only stated here that the words for directions such as up, down, left, right, front, back, inside, and outside that appear or will appear in the text of the present invention are only based on the accompanying drawings of the present invention, and are not specific to the present invention. limited.

As shown in Figure 1, the present invention provides a method for inferring metabolic profiles, including:

S101, subjecting the target sample data to LC-MS technical processing to obtain LC-MS raw data;

S102, performing dimensionality reduction conversion processing on the LC-MS raw data to obtain a two-dimensional matrix, the two-dimensional matrix retaining the retention time, mass-to-charge ratio and ion intensity of the LC-MS raw data;

S103. Input the two-dimensional matrix into the convolutional neural network model to infer the metabolic substance characteristic spectrum of the target sample data.

As shown in Figure 2, preferably, the step of performing dimensionality reduction conversion processing on the LC-MS raw data to obtain a two-dimensional matrix includes;

S201, performing format conversion on the LC-MS raw data;

Converting the LC-MS raw data into .mzml format data, but not limited thereto;

S202, setting a starting retention time, an ending retention time, a retention time interval, a retention time sampling interval, a starting mass-to-charge ratio, an ending mass-to-charge ratio, a mass-to-charge ratio interval, and a mass-to-charge ratio sampling interval.

The range of the retention time interval is the range from the initial retention time to the end retention time, and the mass-to-charge ratio interval is the range from the initial mass-to-charge ratio to the end mass-to-charge ratio;

S203, within the retention time interval and the mass-to-charge ratio interval, using the retention time sampling interval and the mass-to-charge ratio sampling interval as a sliding window, sampling the maximum ion intensity within the retention time interval and the mass-to-charge ratio interval, to obtain a two-dimensional matrix of ionic strength.

The two-dimensional matrix of ionic strength is:

i(t,r)=max{intensity(t,r),...,intensity(t,r+Rgap)...,intensity(t+Tgap,r+Rgap)}, t∈ (T0,Te ), r∈(R0,Re), where t is the retention time, r is the mass-to-charge ratio, intensity is the ionic strength, T0 is the initial retention time, Te is the termination retention time, Tgap is the retention time sampling interval, R0 is The starting mass-to-charge ratio, Re is the ending mass-to-charge ratio, and Rgap is the mass-to-charge ratio sampling interval.

It should be noted that in terms of data preprocessing, the existing deep learning technology obtains two-dimensional mass spectra of each time phase, and carries out substance identification and subsequent analysis based on the mass spectra. The mass spectrum only contains the mass-to-charge ratio and ion intensity information of the substance, plus the phase label of each mass spectrum, which is still a huge three-dimensional data. Therefore, it is necessary to perform a de-redundancy operation to remove a large amount of phase information, so that only a single class of steroids can be processed. However, the present invention innovatively converts the original data in three-dimensional space into a two-dimensional matrix, which can simultaneously retain information such as retention time, mass-to-charge ratio, and ion intensity of the original data. The original data of the serum sample after LC-MS detection is a kind of three-dimensional point cloud data, including three dimensions of retention time, mass-to-charge ratio, and ion intensity. After the dimension reduction transformation is carried out by the method of the present invention, two-dimensional matrix data with retention time and mass-to-charge ratio as axes and ion intensity as values is obtained. It effectively reduces the dimensionality of the original data while retaining the metabolite signal to the greatest extent.

As shown in Figure 3, preferably, the step of inputting the two-dimensional matrix into the convolutional neural network model to infer the metabolic substance profile of the sample includes:

S301. Perform class activation heat map calculation according to the convolutional neural network model to generate a class activation score for each sample.

It should be noted that the class activation score can be expressed as s(t,r), where t is the retention time and r is the mass-to-charge ratio;

S302. Extract a mapping function according to the network structure of the convolutional neural network model.

The mapping function is: t=map1(x), r=map2(y), wherein, t is the retention time, and r is the mass-to-charge ratio;

S303. Map the two-dimensional coordinates of the class activation thermodynamic map to retention time and mass-to-charge ratio according to the mapping function.

Among them, the two-dimensional coordinates (x, y) of the class activation heat map are mapped to retention time (t) and mass-to-charge ratio (r).

S304, performing filtering processing on the class activation score to obtain reserved features;

S305. Screen key metabolites according to the retention characteristics, and perform correlation calculations to infer metabolic markers and metabolic network patterns of the target sample data, and then generate a metabolic profile of the target sample data.

Preferably, the step of filtering the class activation scores to obtain retained features comprises: filtering out molecular features whose class activation scores are less than a first preset threshold and whose ion intensity is less than a second preset threshold, to obtain Preserve features. Specifically, molecular features (t, r) whose class activation scores s(t, r) are smaller than the first threshold or ionic intensity (t, r) is smaller than the second threshold are filtered out to obtain the reserved features [(t1, r1) ,(t2,r2),(t3,r3),...,(tn,rn)].

Specifically, the existing deep learning technology can only classify and extract steroid substances due to the limited types of substances involved. However, the traditional metabolomics technology needs complex data preprocessing to extract the mass spectrum peaks to obtain the metabolite matrix of the sample, and obtain the metabolite characteristic spectrum in a statistically driven manner. According to the characteristics of LC-MS data, the present invention innovatively proposes a mapping function to map the sample features learned by deep learning technology supervision to the original data attributes (retention time, mass-to-charge ratio). For LC-MS data, retention time and mass-to-charge ratio are labels for identifying specific substances. The present invention uses deep learning technology to obtain sample features by calculating the class activation thermodynamic map, and can use the mapping function to infer the specific substances that make up the sample features, thereby further mining the sample feature markers metabolites, metabolic network patterns, and inferring the metabolism of the sample characteristic spectrum.

In addition, the steps to build and train a convolutional neural network model in advance include:

(1) Obtain a data set and divide the data set into a training set, a verification set and a test set, and incorporate data from different sources as an external test set, and use sample attributes as a classification label;

Wherein, the data sets are two-dimensional matrix data obtained through two-dimensional matrix transformation.

(2) construct initial convolutional neural network model, and use training set to carry out model training to described initial convolutional neural network model;

(3) Evaluate the performance of the initial convolutional neural network model after the training in the verification set and the test set, and retrain after adjusting the model structure and hyperparameters if the performance is not good;

(4) The initial convolutional neural network model with the highest accuracy and robustness after training is used as the final convolutional neural network model.

In summary, as shown in Figure 4, the method for inferring metabolic profiles provided by the present invention directly inputs LC-MS raw data, converts and processes the original signal to the greatest extent, and then classifies it using a convolutional neural network model, and Extract features from the classification model to obtain different metabolite patterns in different classifications; using the present invention, the two-dimensional conversion processing of LC-MS raw data can effectively reduce the data size, thereby contributing to subsequent calculations; compared to A large number of signals will be lost when the existing method removes redundancy, and the two-dimensional conversion processing of the LC-MS raw data can preserve the material signal to the greatest extent; the present invention extracts sample attribute-related features from the final convolutional neural network model, It can more effectively evaluate the joint correlation of multiple substances and sample classification, instead of comparing each substance one by one in isolation, so that the metabolic profile of the sample can be inferred more accurately.

As shown in Figure 5, the present invention also provides a metabolic profile inference system 100, including:

LC-MS processing module 1, used for subjecting target sample data to LC-MS technical processing to obtain LC-MS raw data;

Dimensionality reduction conversion processing module 2, for performing dimensionality reduction conversion processing on the LC-MS raw data to obtain a two-dimensional matrix, the two-dimensional matrix retains the retention time, mass-to-charge ratio and ion of the LC-MS raw data strength;

The metabolic profile inference module 3 is used to input the two-dimensional matrix into the convolutional neural network model to infer the metabolic profile of the target sample data.

In the present invention, through the LC-MS processing module 1, first perform LC-MS technology processing on the samples that need to infer the metabolic profile to obtain LC-MS raw data, and then use the dimensionality reduction conversion processing module 2 to convert the LC-MS The MS raw data is subjected to dimensionality reduction conversion processing to obtain a two-dimensional matrix, and finally the two-dimensional matrix is input into the convolutional neural network model through the metabolic profile inference module 3 to infer the metabolic substance profile of the sample. With the present invention, the two-dimensional conversion processing of LC-MS raw data can effectively reduce the data size, thereby contributing to subsequent calculations; compared with the existing methods that will cause a large number of signal loss when removing redundancy, the LC-MS The two-dimensional conversion processing of MS raw data can preserve the material signal to the greatest extent; the present invention extracts sample attribute-related features from the final convolutional neural network model, which can more effectively evaluate the joint correlation between multiple substances and sample classification, and Instead of comparing each substance one by one in isolation, it is possible to more accurately infer the metabolic profile of the sample.

As shown in Figure 6, the dimensionality reduction conversion processing module 2 includes:

A format conversion unit 21, configured to convert the format of the LC-MS raw data;

The parameter setting unit 22 is used to set the initial retention time, termination retention time, retention time interval, retention time sampling interval, initial mass-to-charge ratio, termination mass-to-charge ratio, mass-to-charge ratio interval, and mass-to-charge ratio sampling interval, wherein, The range of the retention time interval is the range from the initial retention time to the end retention time, and the mass-to-charge ratio interval is the range from the initial mass-to-charge ratio to the end mass-to-charge ratio;

The dimensionality reduction sampling unit 23 is used to sample the retention time interval and the mass-to-charge ratio interval using the retention time sampling interval and the mass-to-charge ratio sampling interval as a sliding window within the retention time interval and the mass-to-charge ratio interval. The maximum ionic strength of , to obtain a two-dimensional matrix of ionic strength.

As shown in Figure 7, the metabolic profile inference module 3 includes:

The class activation score acquisition unit 31 is used to calculate the class activation heat map according to the convolutional neural network model, and generate the class activation score s(t, r) of each sample, where t is the retention time and r is the mass charge Compare;

The extraction unit 32 is used to extract the mapping function according to the network structure of the convolutional neural network model: t=map1(x), r=map2(y), wherein, t is the retention time, and r is the mass-to-charge ratio;

A mapping unit 33, configured to map the two-dimensional coordinates of the class activation thermodynamic map to retention time and mass-to-charge ratio according to the mapping function;

A filtering unit 34, configured to filter the class activation scores to obtain reserved features;

Calculation and inference unit 35, configured to screen key metabolites according to the retention characteristics, and perform correlation calculations to deduce the metabolic markers and metabolic network patterns of the target sample data, and then generate the metabolic profile of the target sample data .

Further, the filtering unit is configured to filter out molecular features whose class activation scores are smaller than a first preset threshold and whose ion intensity is smaller than a second preset threshold, so as to obtain reserved features.

It should be noted that the existing deep learning technology can only classify and extract steroid substances due to the limited types of substances involved. However, the traditional metabolomics technology needs complex data preprocessing to extract the mass spectrum peaks to obtain the metabolite matrix of the sample, and obtain the metabolite characteristic spectrum in a statistically driven manner. According to the characteristics of LC-MS data, the present invention innovatively proposes a mapping function to map the sample features learned by deep learning technology supervision to the original data attributes (retention time, mass-to-charge ratio). For LC-MS data, retention time and mass-to-charge ratio are labels for identifying specific substances. The present invention uses deep learning technology to obtain sample features by calculating the class activation thermodynamic map, and can use the mapping function to infer the specific substances that make up the sample features, thereby further mining the sample feature markers metabolites, metabolic network patterns, and inferring the metabolism of the sample characteristic spectrum.

In addition, the metabolic profile inference system 100 also includes a model building block, and the model building block includes:

The data set division unit is used to obtain the data set and divide the data set into a training set, a verification set and a test set, and incorporate data from different sources as an external test set, and use sample attributes as classification labels;

Wherein, the data sets are two-dimensional matrix data obtained by two-dimensional matrix transformation;

Constructing a training unit for constructing an initial convolutional neural network model, and using a training set to perform model training on the initial convolutional neural network model;

The evaluation unit is used to evaluate the performance of the initial convolutional neural network model after training in the verification set and the test set, and retrain after adjusting the model structure and hyperparameters if the performance is not good;

The screening unit is configured to use the initial convolutional neural network model with the highest accuracy and robustness after training as the final convolutional neural network model.

Correspondingly, the present invention also provides a computer device, including a memory, a processor, and computer instructions stored in the memory and operable on the processor. The steps of the above method are implemented when the processor executes the instructions. At the same time, the present invention also provides a storage medium, which stores computer instructions, and when the computer instructions are executed by a processor, the steps of the above method are realized.

In summary, the present invention directly inputs LC-MS raw data, converts and processes the original signal to the greatest extent, and uses the convolutional neural network model to classify, and extracts features from the classification model to obtain different metabolites in different classifications mode; adopt the present invention, the two-dimensional conversion processing to LC-MS raw data, can effectively reduce data size, thereby contribute to follow-up calculation; Compared with the large amount of signal loss that can cause when existing method removes redundancies, to The two-dimensional conversion processing of LC-MS raw data can preserve the material signal to the greatest extent; the present invention extracts sample attribute-related features from the final convolutional neural network model, which can more effectively evaluate the joint correlation between multiple substances and sample classification , instead of comparing each substance one by one in isolation, so that the metabolic profile related to the sample can be inferred more accurately.

The above description is a preferred embodiment of the present invention, and it should be pointed out that for those skilled in the art, without departing from the principle of the present invention, some improvements and modifications can also be made, and these improvements and modifications are also considered Be the protection scope of the present invention.

Claims

A metabolic profile inference method, characterized in that, comprising:

The target sample data is processed by LC-MS technology to obtain LC-MS raw data;

performing a dimensionality reduction conversion process on the LC-MS raw data to obtain a two-dimensional matrix, the two-dimensional matrix retains the retention time, mass-to-charge ratio and ion intensity of the LC-MS raw data;

The two-dimensional matrix is input into the convolutional neural network model to infer the metabolic substance characteristic spectrum of the target sample data.
The method for inferring metabolic profiles according to claim 1, wherein the step of performing dimensionality reduction conversion processing on the LC-MS raw data to obtain a two-dimensional matrix comprises;

Carrying out format conversion of the LC-MS raw data;

Set the initial retention time, termination retention time, retention time interval, retention time sampling interval, initial mass-to-charge ratio, end mass-to-charge ratio, mass-to-charge ratio interval, and mass-to-charge ratio sampling interval, wherein the retention time interval is from From the initial retention time to the range between the termination retention time, the mass-to-charge ratio interval is the range from the initial mass-to-charge ratio to the termination mass-to-charge ratio;

In the retention time interval and the mass-to-charge ratio interval, with the retention time sampling interval and the mass-to-charge ratio sampling interval as a sliding window, sample the maximum ion intensity in the retention time interval and the mass-to-charge ratio interval to obtain ion Intensity 2D matrix.
The metabolic profile inference method according to claim 2, wherein the step of inputting the two-dimensional matrix into a convolutional neural network model to deduce the metabolic profile of the sample comprises:

According to the convolutional neural network model, the class activation heat map is calculated to generate a class activation score s(t, r) for each sample, where t is the retention time and r is the mass-to-charge ratio;

According to the network structure of described convolutional neural network model, extract mapping function: t=map1 (x), r=map2 (y), wherein, t is retention time, and r is mass-to-charge ratio;

mapping the two-dimensional coordinates of the class activation thermodynamic map to retention time and mass-to-charge ratio according to the mapping function;

Filtering the class activation scores to obtain retained features;

Screen key metabolites according to the retention characteristics, and perform correlation calculations to deduce the metabolic markers and metabolic network patterns of the target sample data, and then generate the metabolic profile of the target sample data.
The method for inferring metabolic profiles according to claim 3, wherein the step of filtering the class activation scores to obtain retained features comprises:

Filtering out molecular features whose class activation scores are smaller than a first preset threshold and whose ion intensity is smaller than a second preset threshold, so as to obtain reserved features.
A system for inferring metabolic profiles, characterized by comprising:

The LC-MS processing module is used to process the target sample data with LC-MS technology to obtain LC-MS raw data;

A dimensionality reduction transformation processing module, configured to perform dimensionality reduction transformation processing on the LC-MS raw data to obtain a two-dimensional matrix, the two-dimensional matrix retains the retention time, mass-to-charge ratio and ion intensity of the LC-MS raw data ;

The metabolic profile inference module is used to input the two-dimensional matrix into the convolutional neural network model to infer the metabolic profile of the target sample data.
The metabolic profile inference system according to claim 5, wherein the dimension reduction conversion processing module comprises:

a format conversion unit, configured to convert the format of the LC-MS raw data;

The parameter setting unit is used to set the initial retention time, the termination retention time, the retention time interval, the retention time sampling interval, the initial mass-to-charge ratio, the final mass-to-charge ratio, the mass-to-charge ratio interval, and the mass-to-charge ratio sampling interval, wherein the The range of the retention time interval is the range from the initial retention time to the termination retention time, and the mass-to-charge ratio interval is the range from the initial mass-to-charge ratio to the termination mass-to-charge ratio;

The dimensionality reduction sampling unit is used to sample the retention time interval and the mass-to-charge ratio interval within the retention time interval and the mass-to-charge ratio interval using the retention time sampling interval and the mass-to-charge ratio sampling interval as a sliding window The maximum ionic strength of , to obtain a two-dimensional matrix of ionic strength.
The metabolic profile inference system according to claim 6, wherein the metabolic profile inference module comprises:

The class activation score acquisition unit is used to calculate the class activation heat map according to the convolutional neural network model, and generate the class activation score s(t, r) of each sample, where t is the retention time and r is the mass-to-charge ratio ;

The extraction unit is used to extract the mapping function according to the network structure of the convolutional neural network model: t=map1(x), r=map2(y), where t is the retention time, and r is the mass-to-charge ratio;

a mapping unit, configured to map the two-dimensional coordinates of the class activation thermodynamic map to retention time and mass-to-charge ratio according to the mapping function;

a filtering unit, configured to filter the class activation scores to obtain reserved features;

The calculation and inference unit is used to screen key metabolites according to the retention characteristics, and perform correlation calculations to deduce the metabolic markers and metabolic network patterns of the target sample data, and then generate the metabolic profile of the target sample data.
The metabolic profile inference system according to claim 7, wherein the filtering unit is used to filter out molecular features whose class activation scores are less than a first preset threshold and whose ion intensity is less than a second preset threshold, so as to Get preserved features.
A computer device, comprising a memory, a processor, and computer instructions stored on the memory and operable on the processor, wherein the processor implements any one of claims 1-4 when executing the instructions method steps.
A storage medium, which stores computer instructions, characterized in that, when the computer instructions are executed by a processor, the steps of the method described in any one of claims 1-4 are implemented.