A chromatographic mass spectrometry automatic integration method, system, device and medium

Abstract

The present invention relates to a chromatogram mass spectrometry automatic integration method, system, equipment, and medium, and belongs to the technical field of proteomics and metabolomics analysis. The original chromatogram mass spectrometry is first converted to obtain a standardized spectrogram image, and then the standardized spectrogram image is preliminarily identified using a peak recognition algorithm to obtain a candidate peak result, and the standardized spectrogram image is predicted using a trained deep learning model to obtain a spectrogram classification result, and finally the candidate peak results are screened based on the spectrogram classification result, and the peak results predicted as noise in the candidate peak results are removed to obtain an automatic integration result, and the acquired chromatogram mass spectrometry image is predicted by a deep learning model, which can replace manual discrimination of the peak shape of the chromatogram mass spectrometry, and the method has high accuracy and high specificity, effectively reduces the false positives of conventional automatic integration methods, and can fully and efficiently realize the automatic replacement of manual integration.

Description

Technical Field

[0001] The present invention relates to the technical field of proteomics and metabolomics analysis, and in particular to a chromatographic mass spectrometry automatic integration method, system, equipment and medium based on a deep learning model. Background Art

[0002] Proteomics and metabolomics are two new disciplines developed in the field of systems biology after genomics. Proteomics studies the composition, activity patterns and protein-protein interactions of proteins in cells from the overall level of proteins, while metabolomics explores the metabolic mechanism of the entire organism by detecting the changes in metabolites in the body after exogenous stimulation or genetic modification.

[0003] As a combined technology, chromatography-tandem mass spectrometry combines the efficient separation capability of chromatography with the precise molecular weight and intensity information of the molecules to be detected collected by mass spectrometry. This technology has the advantages of high sensitivity, strong selectivity and high throughput, and can obtain rich qualitative and quantitative information on proteins and metabolites. Therefore, it has significant advantages in the field of proteomics and metabolomics analysis.

[0004] However, due to the diversity of chromatographic mass spectrometry peak results, the integration results provided by the instrument software often have the disadvantages of high false positives and inaccurate peak shape judgment, which requires manual verification. This is not only time-consuming and labor-intensive, but also leads to differences in the judgment of peak results by different people, which has an adverse impact on the consistency of metabolomics and proteomics results.

[0005] Based on this, there is an urgent need for a chromatographic mass spectrometry automatic integration method, system, equipment, and medium that can automatically perform peak shape discrimination. Summary of the invention

[0006] The purpose of the present invention is to provide a method, system, device and medium for automatic integration of chromatogram mass spectrometry, which uses a deep learning model to automatically identify the peak shape of the chromatogram mass spectrometry, and can obtain quantitative results of the chromatogram mass spectrometry peaks with high accuracy and efficiency.

[0007] To achieve the above object, the present invention provides the following solutions:

[0008] A chromatogram mass spectrometry automatic integration method, the automatic integration method comprising:

[0009] Convert the original spectrum of the chromatogram and mass spectrum to obtain a standardized spectrum image;

[0010] Performing preliminary peak recognition on the standardized spectrum image using a peak recognition algorithm to obtain candidate peak results;

[0011] The standardized spectrogram image is predicted using the trained deep learning model to obtain a spectrogram classification result; the spectrogram classification result includes the type of each peak shape in the standardized spectrogram image; the type includes noise and normal;

[0012] The candidate peak results are screened based on the spectrum classification results, and the peak results predicted to be noise in the candidate peak results are removed to obtain automatic integration results.

[0013] A chromatogram mass spectrometer automatic integration system, the automatic integration system comprising:

[0014] A conversion module is used to convert the original spectrum of the chromatogram and mass spectrum to obtain a standardized spectrum image;

[0015] A preliminary identification module is used to perform preliminary peak identification on the standardized spectrum image using a peak identification algorithm to obtain candidate peak results;

[0016] A prediction module, used to predict the standardized spectrogram image using a trained deep learning model to obtain a spectrogram classification result; the spectrogram classification result includes the type of each peak in the standardized spectrogram image; the type includes noise and normal;

[0017] A screening module is used to screen the candidate peak results based on the spectrum classification results, remove the peak results whose predicted results are noise in the candidate peak results, and obtain automatic integration results.

[0018] A chromatographic mass spectrometry automatic integration device, comprising:

[0019] Processor; and

[0020] a memory storing computer readable program instructions,

[0021] Wherein, the above automatic integration method is executed when the computer-readable program instructions are executed by the processor.

[0022] A computer-readable storage medium stores a computer program, which implements the steps of the automatic integration method when executed by a processor.

[0023] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0024] The present invention is used to provide a method, system, equipment and medium for automatic integration of chromatogram mass spectrometry. The original chromatogram mass spectrometry spectrum is first converted to obtain a standardized spectrum image, and then a peak recognition algorithm is used to perform preliminary peak recognition on the standardized spectrum image to obtain candidate peak results. At the same time, a trained deep learning model is used to predict the standardized spectrum image to obtain a spectrum classification result. Finally, the candidate peak results are screened based on the spectrum classification result, and peak results predicted as noise in the candidate peak results are removed to obtain automatic integration results. By performing deep learning model prediction on the acquired chromatogram mass spectrometry image, manual peak shape discrimination of the chromatogram mass spectrum can be replaced. The method has high accuracy and high specificity, effectively reduces the false positives of conventional automatic integration methods, and is fully and efficiently capable of realizing automated replacement of manual integration. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0026] Figure 1 A method flow chart of the automatic integration method provided in Example 1 of the present invention;

[0027] Figure 2 A method principle diagram of the automatic integration method provided in Example 1 of the present invention;

[0028] Figure 3 A schematic diagram of the network structure of the deep learning model provided in Example 1 of the present invention;

[0029] Figure 4 This is a system block diagram of the automatic integration system provided in Example 2 of the present invention. DETAILED DESCRIPTION

[0030] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0031] The purpose of the present invention is to provide a method, system, device and medium for automatic integration of chromatogram mass spectrometry, which uses a deep learning model to automatically identify the peak shape of the chromatogram mass spectrometry, and can obtain quantitative results of the chromatogram mass spectrometry peaks with high accuracy and efficiency.

[0032] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0033] Embodiment 1:

[0034] This embodiment is used to provide a chromatographic mass spectrometry automatic integration method based on a deep learning model, such as Figure 1 and Figure 2 As shown, the automatic integration method includes:

[0035] S1: Convert the original spectrum of chromatogram and mass spectrum to obtain a standardized spectrum image;

[0036] Specifically, the original chromatographic mass spectra may include: gas chromatography-single quadrupole mass spectrometry, gas chromatography-triple quadrupole mass spectrometry, gas chromatography-time-of-flight mass spectrometry, liquid chromatography-triple quadrupole mass spectrometry, liquid chromatography-time-of-flight mass spectrometry and liquid chromatography-orbitrap mass spectrometry.

[0037] S1 may include: for each mass spectrum channel of the original spectrum of the chromatogram mass spectrum, a spectrum is drawn with time as the horizontal coordinate and mass spectrum signal intensity as the vertical coordinate, and the spectra of all mass spectrum channels constitute a standardized spectrum picture. That is, after the original spectrum of the chromatogram mass spectrum is converted into a standardized spectrum picture in this embodiment, the obtained standardized spectrum picture includes multiple spectra, each spectrum corresponds to the data of a mass spectrum channel, the horizontal coordinate is time, and the vertical coordinate is the mass spectrum signal intensity.

[0038] As an optional implementation, before S1, this embodiment can also convert the data format of the chromatogram mass spectrum original spectrum into a universal mass spectrum data format, such as mzML, mzXML, mgf, etc., through the instrument's own software or MSConvert (version may be V3.0.19243), and use the converted spectrum with the universal mass spectrum data format as the chromatogram mass spectrum original spectrum to execute S1. By converting the data format of the chromatogram mass spectrum original spectrum, the method of this embodiment can be applied to chromatogram mass spectrum original spectra in different data formats provided by different instrument manufacturers, and has a wide range of applications.

[0039] Before S1, this embodiment can also read the original chromatographic mass spectrum and extract the spectrum information therein, which may include the parent ion mass number, the daughter ion mass number, the retention time, the signal intensity, etc., and execute S1 based on the spectrum information.

[0040] S2: using a peak recognition algorithm to perform preliminary peak recognition on the standardized spectrum image to obtain candidate peak results;

[0041] The peak identification algorithm can be a multi-parameter, multi-method integration method, which includes the Centwave (wavelet filtering algorithm) and MatchFilter (matching filtering algorithm) of XCMS, and uses multiple calculation parameters for peak identification to prevent partial peak loss, such as multiple peak widths with peak width parameters between 3-20s. There are a large number of false positive peaks in the candidate peak results obtained by the peak identification algorithm, which need to be filtered and removed using a subsequent trained deep learning model.

[0042] As an optional implementation, before using the trained deep learning model to screen the candidate peak results, the automatic integration method of this embodiment also includes: deduplicating and screening the initial candidate peak results obtained by the multi-parameter, multi-method integration method according to parameters such as peak height, signal-to-noise ratio and peak width, to obtain the peak results after preliminary screening, and use the peak results after preliminary screening as new candidate peak results, so as to perform preliminary screening on the candidate peak results first, thereby improving the screening efficiency.

[0043] S3: using the trained deep learning model to predict the standardized spectrum image to obtain a spectrum classification result; the spectrum classification result includes the type of each peak in the standardized spectrum image; the type includes noise and normal;

[0044] Before using the trained deep learning model to predict the standardized spectrogram image and obtain the spectrogram classification result, the automatic integration method of this embodiment also includes the step of training the deep learning model to obtain the trained deep learning model, which may include:

[0045] (1) obtaining a data set; the data set includes a plurality of standardized spectrum images for training and a label corresponding to each standardized spectrum image for training; the label is the type of all peak shapes in the standardized spectrum image for training;

[0046] In this embodiment, a manual labeling method is used to label and classify the peak shape of each training standardized spectrum image to obtain the label of the training standardized spectrum image. All training standardized spectrum images and the label of each training standardized spectrum image constitute a data set. Specifically, the data set used in the training process can be a manually labeled actual project standardized spectrum image, the number of which is 1.3 million.

[0047] (2) Build a deep learning model;

[0048] The deep learning model of this embodiment can be composed of the following deep learning basic frameworks: convolutional neural layer (CNN), long short-term memory layer (LSTM) and attention mechanism (Transformer). The attention mechanism refers to the Transformer model, which includes an encoder and a decoder constructed by multi-head attention (Multi-Head Attention).

[0049] Specifically, Figure 3 As shown, the deep learning model of this embodiment may include an input layer, multiple convolutional neural layers, long short-term memory layers, fully connected layers, and output layers connected in sequence, the activation function of each intermediate layer is a linear rectifier function ReLU, and the activation function of the output layer is SoftMax. In order to reduce the risk of overfitting, a regularized normalization layer BatchNorm and a dropout layer are added between each layer. That is, each convolutional neural layer, long short-term memory layer, and fully connected layer includes a ReLU activation function, a regularized normalization layer, and a dropout layer, and the output layer includes a SoftMax activation function.

[0050] (3) Divide the data set into training set and validation set;

[0051] 70% of the data in the dataset can be used as the training set, and the remaining 30% of the data can be used as the test set.

[0052] (4) Using the training set as input, the deep learning model is trained to obtain a trained model;

[0053] For the standardized spectrogram images used for training, their essence is a two-dimensional matrix. All the standardized spectrogram images used for training are converted to the same dimension through linear interpolation method, and then normalized and standardized as the input matrix of the deep learning model. The input matrix is ​​passed through continuous N (such as Figure 3 As shown in the figure, N is 3) one-dimensional convolutional neural layers CNN, and the number of nodes in each layer can be 8, 64, and 512 respectively, so as to increase the dimension from the initial 64*1 to 64*512, and then extract the long and short features through a long short memory layer LSTM layer and reduce the dimension to 64 layers, and output the final prediction result through the fully connected layer. The specific values ​​of the number of nodes and dimensions mentioned above are only examples, and can be changed according to user needs in actual applications.

[0054] During the training process, the loss function of the model is cross entropy CrossEntropyLoss, the number of training Epochs is 200, the initial training learning rate is 0.005, and it decreases with the increase of training times.

[0055] (5) Adjust the hyperparameters of the trained model to obtain the optimized model;

[0056] In this embodiment, the adjustable model hyperparameters mainly include learning rate, number of training times, number of model layers, number of nodes in the middle layer, etc. The model is optimized by optimizing hyperparameters such as number of training times, number of nodes, and batch size to obtain an optimized model.

[0057] (6) Determine whether the preset number of iterations has been reached;

[0058] (7) If yes, use the test set as input, calculate the prediction effects of all optimized models, and select the optimized model with the best prediction effect as the trained deep learning model;

[0059] The final model was obtained after evaluation using the test set, and the model with the best effect on the test set was selected as the trained deep learning model. During the experiment, the final accuracy of the model was 93% and the specificity was 90%.

[0060] (8) If not, the optimized model is used as the deep learning model in the next iteration, and the process returns to the step of “using the training set as input to train the deep learning model to obtain the trained model”.

[0061] S4: Screening the candidate peak results based on the spectrum classification result, removing the peak results whose predicted results are noises from the candidate peak results, and obtaining an automatic integration result.

[0062] After manual verification, the overlap rate between the results of the automatic integration method of this embodiment and the manual method is more than 98%, which can meet the needs of various subsequent analysis methods.

[0063] The automatic integration method provided in this embodiment can be divided into three parts. The first part: converting the original spectrum of the chromatographic mass spectrometer into a standardized spectrum picture, and using a multi-parameter, multi-method integration method to obtain the candidate peak results. The second part: the construction and training part of the deep learning model, the training part includes manually annotating and classifying the standardized spectrum pictures, and dividing them into a training set and a test set. The deep learning model is first trained with the data of the training set, and the model hyperparameters are adjusted. Then the model adjustment results are evaluated using the test set data, and the model with the best effect on the test set is selected as the trained deep learning model. The third part: based on the trained deep learning model, the candidate peak results are screened, specifically using the trained deep learning model to predict the standardized spectrum pictures, obtain the spectrum classification results, and then use the spectrum classification results to screen the candidate peak results, filter and remove the peak results whose prediction results are noise, and obtain the final automatic integration results. Through this method, instead of manually realizing the accurate judgment of the peak shape, it can effectively solve the shortcomings of high false positives and inaccurate peak shape judgment of conventional chromatographic mass spectrometry automatic integration, thereby obtaining the quantitative results of the chromatographic mass spectrometry peaks with high accuracy and efficiency.

[0064] This embodiment creatively provides a chromatographic mass spectrometry automatic integration method based on a deep learning model. By performing deep learning model prediction on the acquired chromatographic mass spectrometry image, it can replace manual discrimination of the peak shape of the chromatographic mass spectrometry. This method has high accuracy and high specificity, effectively reduces the false positives of conventional automatic integration methods, and is fully capable of efficiently realizing automated replacement of manual integration.

[0065] Embodiment 2:

[0066] This embodiment is used to provide a chromatogram mass spectrometer automatic integration system, such as Figure 4 As shown, the automatic integration system comprises:

[0067] The conversion module M1 is used to convert the original spectrum of the chromatogram and mass spectrum to obtain a standardized spectrum image;

[0068] A preliminary identification module M2 is used to perform preliminary peak identification on the standardized spectrum image using a peak identification algorithm to obtain candidate peak results;

[0069] The prediction module M3 is used to predict the standardized spectrogram image using the trained deep learning model to obtain a spectrogram classification result; the spectrogram classification result includes the type of each peak in the standardized spectrogram image; the type includes noise and normal;

[0070] The screening module M4 is used to screen the candidate peak results based on the spectrum classification results, remove the peak results whose predicted results are noises in the candidate peak results, and obtain automatic integration results.

[0071] Embodiment 3:

[0072] This embodiment is used to provide a chromatogram mass spectrometer automatic integration device, including:

[0073] Processor; and

[0074] a memory storing computer readable program instructions,

[0075] Wherein, when the computer-readable program instructions are executed by the processor, the automatic integration method as described in Example 1 is executed.

[0076] Embodiment 4:

[0077] This embodiment is used to provide a computer-readable storage medium, on which a computer program is stored, characterized in that when the program is executed by a processor, the steps of the automatic integration method described in Example 1 are implemented.

[0078] Although the present invention is described herein in conjunction with various embodiments, in the process of implementing the claimed invention, those skilled in the art may understand and implement other variations of the disclosed embodiments by viewing the drawings, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other components or steps, and "one" or "an" does not exclude multiple situations. A single processor or other unit may implement several functions listed in the claims. Certain measures are recorded in different dependent claims, but this does not mean that these measures cannot be combined to produce good results.

[0079] Although the present invention has been described in conjunction with specific features and embodiments thereof, it is apparent that various modifications and combinations may be made thereto without departing from the spirit and scope of the present invention. Accordingly, this specification and the accompanying drawings are merely exemplary illustrations of the present invention as defined by the appended claims and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present invention. Obviously, those skilled in the art may make various modifications and variations to the present invention without departing from the spirit and scope of the present invention. Thus, the present invention is intended to include such modifications and variations if they fall within the scope of the claims of the present invention and their equivalents.

Claims

1. A method for automatic integration of chromatogram mass spectrometry, characterized in that: The automatic integration method comprises: Convert the original chromatographic mass spectrometry to obtain a standardized spectrum image; Performing preliminary peak recognition on the standardized spectrum image using a peak recognition algorithm to obtain candidate peak results; The standardized spectrogram image is predicted using the trained deep learning model to obtain a spectrogram classification result; the spectrogram classification result includes the type of each peak shape in the standardized spectrogram image; the type includes noise and normal; The candidate peak results are screened based on the spectrum classification results, and the peak results predicted to be noise in the candidate peak results are removed to obtain automatic integration results; in, Before obtaining the standardized spectrum image, the original spectrum of the chromatogram mass spectrum is read and the spectrum information therein is extracted, and the spectrum information includes the spectrum information of the parent ion mass number, the daughter ion mass number, the retention time, and the signal intensity; The converting of the original chromatographic mass spectrometer to obtain a standardized spectrum image specifically includes: For each mass spectrum channel of the original spectrum of the chromatographic mass spectrum, a spectrum is plotted with time as the horizontal axis and mass spectrum signal intensity as the vertical axis, and the spectra of all the mass spectrum channels constitute a standardized spectrum picture; The peak identification algorithm is a multi-parameter, multi-method integration method, and the multi-parameter, multi-method integration method includes centwave and matchfilter; Before screening the candidate peak results based on the spectrum classification results, the automatic integration method further includes: De-duplication and screening of the candidate peak results are performed according to peak height, signal-to-noise ratio and peak width to obtain preliminary screened peak results, and the preliminary screened peak results are used as new candidate peak results; Before using the trained deep learning model to predict the standardized spectrogram image to obtain the spectrogram classification result, the automatic integration method also includes training the deep learning model to obtain the trained deep learning model, specifically including: Acquire a data set; the data set includes a plurality of standardized spectrum images for training and a label corresponding to each of the standardized spectrum images for training; the label is the type of all peak shapes in the standardized spectrum images for training; Build deep learning models; Dividing the data set into a training set and a validation set; Using the training set as input, training the deep learning model to obtain a trained model; Adjusting the hyperparameters of the trained model to obtain an optimized model; Determine whether the preset number of iterations has been reached; If yes, then taking the validation set as input, calculating the prediction effects of all the optimized models, and selecting the optimized model with the best prediction effect as the trained deep learning model; If not, the optimized model is used as the deep learning model in the next iteration, and the process returns to the step of "using the training set as input to train the deep learning model to obtain a trained model"; The deep learning model includes an input layer, a plurality of convolutional neural layers, a long short-term memory layer, a fully connected layer and an output layer connected in sequence; each of the convolutional neural layer, the long short-term memory layer and the fully connected layer includes a ReLU activation function, a regularization layer and a dropout layer, and the output layer includes a SoftMax activation function; During the training process, the loss function of the model is cross entropy CrossEntropyLoss, the number of training Epochs is 200, the initial training learning rate is 0.005, and it decreases with the increase of training times.

2. The automatic integration method according to claim 1, characterized in that: The chromatographic mass spectrometry original spectra include: gas chromatography-single quadrupole mass spectrometry, gas chromatography-triple quadrupole mass spectrometry, gas chromatography-time-of-flight mass spectrometry, liquid chromatography-triple quadrupole mass spectrometry, liquid chromatography-time-of-flight mass spectrometry and liquid chromatography-orbitrap mass spectrometry.

3. A chromatographic mass spectrometry automatic integration device, characterized in that: include: processor; as well as a memory storing computer readable program instructions, Wherein, when the computer-readable program instructions are executed by the processor, the automatic integration method as claimed in any one of claims 1 to 2 is executed.

4. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the automatic integration method according to any one of claims 1 to 2 are implemented.

Images