Peptide fragment sequencing method, device, equipment, storage medium and program product

Abstract

The invention provides a peptide fragment sequencing method, device and equipment, a storage medium and a program product, and relates to the technical field of artificial intelligence, and the method comprises the following steps: obtaining mass spectrum data of a peptide fragment to be sequenced; inputting the mass spectrum data into a peptide fragment sequencing model to obtain a peptide fragment sequencing result output by the peptide fragment sequencing model; wherein the peptide fragment sequencing model is obtained by training an initial model constructed based on a TSARseqNovo algorithm on the basis of the sample mass spectrum data and a sample peptide fragment sequencing result corresponding to the sample mass spectrum data. Through the mode, the analysis precision of mass spectrum information can be remarkably improved, the reasoning efficiency of the model is improved, and the quality and accuracy of a peptide fragment sequencing result can be effectively guaranteed.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a peptide sequencing method, device, equipment, storage medium and program product. Background Art

[0002] In mass spectrometry-based proteomics technology, peptide identification plays a vital role in analyzing the protein composition of complex biological samples. The processing steps of peptide identification include collecting mass spectrometry images of peptides for detailed analysis and pairing them according to key data (such as mass number, charge, retention time, etc.) to infer the amino acid sequence that constitutes them. At present, the main technologies for peptide identification include library search technology, which compares and matches the observed mass spectrometry information with the pre-stored peptide information in the existing database to achieve the purpose of identifying peptides. The limitation of this method is that when identifying antibodies, new antigens in human leukocyte antigens, or new proteins and peptides that are not listed in the current genomics database, this method cannot effectively achieve the identification of related peptides.

[0003] Another mainstream method is the peptide de novo sequencing method, which refers to the analysis process of obtaining the peptide amino acid sequence through tandem mass spectrometry without the help of a sequence database. In the early days, non-deep learning methods were mainly used, mainly including heuristic search, dynamic programming and other methods to evaluate the degree of match between peptides and corresponding spectral observations. Subsequently, the Novo algorithm combined the decision tree in the scoring function to enhance the Novo model's ability to recognize peptides. With the explosive development of deep learning methods, some researchers proposed the DeepNovo model, which combines convolutional neural networks with long and short-term memory networks, significantly improving the peptide de novo sequencing ability of the DeepNovo model. On this basis, the PointNovo model improves its feature extraction ability for high-resolution mass spectrometry data by adopting an order-invariant network architecture, further improving the peptide de novo sequencing ability. The CasaNovo model is based on the Transformer framework, which directly maps the peaks observed from the mass spectrometry data to the amino acid sequence in the feature space, greatly improving the recognition and sequencing performance of peptides.

[0004] However, as the most commonly used model at present, the CasaNovo model also has some shortcomings: in terms of inference objects, the CasaNovo model mainly relies on the historical information of peptides for inference and ignores the information of individual amino acids. Although this strategy improves its prediction performance for peptides, the accuracy decreases when analyzing single amino acid information; in addition, the CasaNovo model does not fully utilize the spatial arrangement correlation of spectral peaks when analyzing mass-to-charge ratio and intensity information. These shortcomings significantly affect the CasaNovo model's ability to extract global and local features of mass spectrometry information; in terms of inference speed, the CasaNovo model can only handle the prediction analysis of one amino acid in a single inference process, which significantly slows down the reasoning process.

[0005] Therefore, the accuracy of existing peptide sequencing methods is not high. Summary of the invention

[0006] The present invention provides a peptide sequencing method, device, equipment, storage medium and program product, which are used to solve the defect of low accuracy of peptide sequencing methods in the prior art.

[0007] The present invention provides a peptide sequencing method, comprising: obtaining mass spectrum data of a peptide to be sequenced; inputting the mass spectrum data into a peptide sequencing model to obtain a peptide sequencing result output by the peptide sequencing model; wherein the peptide sequencing model is obtained by training an initial model constructed based on a TSARseqNovo algorithm based on sample mass spectrum data and sample peptide sequencing results corresponding to the sample mass spectrum data.

[0008] According to a peptide sequencing method provided by the present invention, the peptide sequencing model includes a mass spectrum peak information fusion encoding module, a semi-autoregressive decoding module and a mask optimization decoding module; the mass spectrum peak information fusion encoding module is used to extract features from mass spectrum data to obtain mass spectrum peak fusion embedded features; the semi-autoregressive decoding module is used to perform initial prediction based on the mass spectrum peak fusion embedded features to obtain amino acid prediction results of the peptide to be sequenced; the mask optimization decoding module is used to perform optimized prediction based on the amino acid prediction results to obtain peptide sequencing results.

[0009] According to a peptide sequencing method provided by the present invention, a mass spectrum peak information fusion coding module includes a mass spectrum peak coding module, a local and global mass spectrum feature coding module, a mass spectrum peak fusion coding module and multiple encoders; the mass spectrum peak coding module is used to perform feature coding on mass spectrum data to obtain mass spectrum peak embedded coding; the local and global mass spectrum feature coding module is used to perform feature coding on mass spectrum data to obtain local and global spectrum feature embedded coding; the mass spectrum peak fusion coding module is used to perform feature fusion on mass spectrum peak embedded coding and local and global spectrum feature embedded coding to obtain mass spectrum peak fusion coding; all encoders are used to generate mass spectrum peak fusion embedded features based on mass spectrum peak fusion coding.

[0010] According to a peptide sequencing method provided by the present invention, an initial prediction is performed based on mass spectrometry peak fusion embedding features to obtain an amino acid prediction result of the peptide to be sequenced, including: generating amino acid coding information of the peptide to be sequenced; performing an initial prediction based on the amino acid coding information and mass spectrometry peak fusion embedding features to obtain an amino acid prediction result of the peptide to be sequenced.

[0011] According to a peptide sequencing method provided by the present invention, the amino acid prediction result includes multiple amino acid identifiers of the peptide to be sequenced and the prediction confidence of each amino acid identifier; optimizing prediction is performed based on the amino acid prediction result to obtain the peptide sequencing result, including: determining the target amino acid identifier based on the amino acid prediction result; the target amino acid identifier is an amino acid identifier with a prediction confidence less than a preset threshold; optimizing prediction is performed based on the target amino acid identifier and mass spectrum peak fusion embedding features to generate the peptide sequencing result.

[0012] According to a peptide sequencing method provided by the present invention, a peptide sequencing model is trained using a preset loss function; wherein the preset loss function is determined based on a first loss function and a second loss function, the first loss function is a cross entropy loss function of a semi-autoregressive decoding module, and the second loss function is a cross entropy loss function of a mask optimized decoding module.

[0013] The present invention also provides a peptide sequencing device, comprising: an acquisition module, used to obtain mass spectrum data of a peptide to be sequenced; a sequencing module, used to input the mass spectrum data into a peptide sequencing model to obtain a peptide sequencing result output by the peptide sequencing model; wherein the peptide sequencing model is obtained by training an initial model constructed based on a TSARseqNovo algorithm based on sample mass spectrum data and sample peptide sequencing results corresponding to the sample mass spectrum data.

[0014] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, any of the above-mentioned peptide sequencing methods is implemented.

[0015] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, any of the peptide sequencing methods described above is implemented.

[0016] The present invention also provides a computer program product, comprising a computer program, which implements any of the above-mentioned peptide sequencing methods when executed by a processor.

[0017] The peptide sequencing method, apparatus, device, storage medium and program product provided by the present invention use a peptide sequencing model based on the TSARseqNovo algorithm to analyze and predict the mass spectrometry data of the peptide to be sequenced to obtain peptide sequencing results. Compared with the existing CasaNovo model, the peptide sequencing model based on the TSARseqNovo algorithm has better feature encoding ability and reasoning ability, can significantly improve the analysis accuracy of mass spectrometry information, improve the reasoning efficiency of the model, and effectively ensure the quality and accuracy of the peptide sequencing results. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are 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.

[0019] Figure 1 This is one of the schematic diagrams of the peptide sequencing method provided by the present invention.

[0020] Figure 2 This is the second schematic diagram of the process of the peptide sequencing method provided by the present invention.

[0021] Figure 3 It is one of the schematic diagrams of the parallel reasoning of the semi-autoregressive decoding module provided by the present invention.

[0022] Figure 4 This is the second schematic diagram of the parallel reasoning of the semi-autoregressive decoding module provided by the present invention.

[0023] Figure 5 It is a schematic diagram of the processing flow of the semi-autoregressive decoding module provided by the present invention.

[0024] Figure 6 This is one of the schematic diagrams of the evaluation results of different models provided by the present invention on nine species data sets.

[0025] Figure 7 This is the second schematic diagram of the evaluation results of different models provided by the present invention on nine species data sets.

[0026] Figure 8 This is the third schematic diagram of the evaluation results of different models provided by the present invention on nine species data sets.

[0027] Fig. 9 This is one of the schematic diagrams of the evaluation results of different models provided by the present invention on the merged high-resolution dataset.

[0028] Fig.10This is the second schematic diagram of the evaluation results of different models provided by the present invention on the merged high-resolution data set.

[0029] Fig.11 This is the third schematic diagram of the evaluation results of different models provided by the present invention on the merged high-resolution data set.

[0030] Fig.12 It is a schematic diagram of the structure of the peptide sequencing device provided by the present invention.

[0031] Fig.13 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0032] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are 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.

[0033] See also Figure 1 , Figure 1 This is one of the flow charts of the peptide sequencing method provided by the present invention. In this embodiment, the peptide sequencing method includes steps S110 to S120, and each step is specifically as follows: S110: Obtaining mass spectrometry data of the peptide to be sequenced.

[0034] S120: inputting the mass spectrometry data into the peptide sequencing model to obtain the peptide sequencing results output by the peptide sequencing model.

[0035] Among them, the peptide sequencing model is obtained by training the initial model constructed based on the TSARseqNovo algorithm based on the sample mass spectrometry data and the sample peptide sequencing results corresponding to the sample mass spectrometry data.

[0036] The peptide sequencing method provided in this embodiment uses a peptide sequencing model based on the TSARseqNovo algorithm to analyze and predict the mass spectrometry data of the peptide to be sequenced to obtain the peptide sequencing results. Compared with the existing CasaNovo model, the peptide sequencing model based on the TSARseqNovo algorithm has better feature encoding ability and reasoning ability, can significantly improve the analysis accuracy of mass spectrometry information, improve the reasoning efficiency of the model, and effectively ensure the quality and accuracy of the peptide sequencing results.

[0037] In some embodiments, the peptide sequencing model includes a mass spectrum peak information fusion encoding module, a semi-autoregressive decoding module and a mask optimization decoding module; the mass spectrum peak information fusion encoding module is used to extract features from mass spectrum data and obtain mass spectrum peak fusion embedded features; the semi-autoregressive decoding module is used to perform initial predictions based on the mass spectrum peak fusion embedded features to obtain amino acid prediction results of the peptide to be sequenced; the mask optimization decoding module is used to perform optimized predictions based on the amino acid prediction results to obtain peptide sequencing results.

[0038] See also Figure 2 , Figure 2 This is the second schematic diagram of the process of the peptide sequencing method provided by the present invention.

[0039] like Figure 2 As shown, this embodiment adopts a peptide sequencing model based on the TSARseqNovo algorithm, which includes a mass spectrometry peak information fusion encoding (MS PeakFusion Encoder) module, a semi-autoregressive decoding (Semi-autoregressive, SAR) module and a masking optimization (Masking Refinement, MR) decoding module.

[0040] Among them, the mass spectrum peak information fusion encoding module is used to extract features from mass spectrum data and obtain mass spectrum peak fusion embedded features.

[0041] Specifically, the mass spectrum peak information fusion encoding module is mainly used to calculate the mass spectrum peak information fusion embedding features: in this module, the mass spectrum data is first encoded, and the mass spectrum peak encoding and the local and global mass spectrum feature encoding are comprehensively considered; then the feature encoding is fused to obtain the mass spectrum peak fusion encoding; then the mass spectrum peak fusion encoding is input into the standard Transformer encoder to obtain the mass spectrum peak fusion embedding features.

[0042] The semi-autoregressive decoding module is used to make initial predictions based on the mass spectrometry peak fusion embedding features to obtain the amino acid prediction results of the peptide to be sequenced.

[0043] Specifically, the semi-autoregressive decoding module can encode the amino acid information, and after performing a bitwise addition operation on the position information and mass information of the amino acid of the peptide to be sequenced, the amino acid coding information corresponding to the peptide to be sequenced is obtained, wherein the identifier of the initial identification position of the amino acid of each peptide is set to [b], the end identifier is set to [e], and the length of each peptide is 100, that is, the longest encoding length of the amino acid is 100; after obtaining the amino acid coding information, the semi-autoregressive decoding module can use the amino acid coding information and the mass spectrum peak fusion embedding feature as its input, and obtain a new set of amino acid identifiers and their confidences after calculation by the semi-autoregressive decoding module; replace the current amino acid identifier with the new amino acid prediction result and re-encode it, and repeat the above process. The process is cyclically predicted until a predetermined stop identifier [e] is encountered or the number of amino acids reaches 100. In this process, for each set of predicted results, the amino acid identifiers with a prediction confidence less than a preset threshold are screened out and marked as [m] to obtain the amino acid prediction results of the peptide to be sequenced.

[0044] The mask optimization decoding module is used to optimize the prediction based on the amino acid prediction results to obtain the peptide sequencing results.

[0045] Specifically, the mask optimization decoding module can perform secondary predictions based on the amino identifiers corresponding to the positions marked as [m] and the mass spectrum peak fusion embedding features until the prediction confidence of all predicted amino identifiers is greater than or equal to the preset threshold, thereby generating the final peptide sequencing results and optimizing the amino acids to improve the overall peptide prediction accuracy.

[0046] In some embodiments, the mass spectrum peak information fusion coding module includes a mass spectrum peak coding module, a local and global mass spectrum feature coding module, a mass spectrum peak fusion coding module and multiple encoders; the mass spectrum peak coding module is used to feature encode the mass spectrum data to obtain the mass spectrum peak embedded coding; the local and global mass spectrum feature coding module is used to feature encode the mass spectrum data to obtain the local and global spectrum feature embedded coding; the mass spectrum peak fusion coding module is used to feature fuse the mass spectrum peak embedded coding and the local and global spectrum feature embedded coding to obtain the mass spectrum peak fusion coding; all encoders are used to generate mass spectrum peak fusion embedded features based on the mass spectrum peak fusion coding.

[0047] In the model architecture based on the Transformer framework, the embedded coding of data can reflect the subtle changes between different spectral peaks and the difference information between different amino acids, so it plays a very important role in the subsequent prediction and analysis of peptides. The CasaNovo model uses sine-cosine coding technology to convert the mass-to-charge ratio and intensity of mass spectrometry data into high-dimensional vector embedding. Although this embedding method can reflect the differences between different spectral peaks, it fails to consider the correlation between each peak. Therefore, the peptide sequencing model based on the TSARseqNovo algorithm introduces a new embedded coding method. The implementation of this embedded coding method depends on three main components: mass spectrometry peak coding module, local and global mass spectrometry feature coding module (i.e., local-global mass spectrometry feature coding module), and mass spectrometry peak fusion coding module.

[0048] Specifically, the mass spectrum peak information fusion coding module includes a mass spectrum peak coding module, a local and global mass spectrum feature coding module, a mass spectrum peak fusion coding module and multiple encoders.

[0049] The mass spectrum peak encoding module uses the sine-cosine encoding technology in the CasaNovo algorithm to feature encode the mass spectrum data, convert the peak information of the mass spectrum data into a high-dimensional vector, and obtain the mass spectrum peak embedding code.

[0050] The embedding formula for each peak is as follows: In the above formula, represents the mass-to-charge ratio of the ith peak, Represents the dimension size of the entire spectrum peak embedding, represents the minimum change in the sine-cosine wavelength, Represents the maximum value of the change in sine-cosine wavelength.

[0051] This embedded coding method can convert the mass-to-charge ratio and intensity in mass spectrometry data into vector information in a high-dimensional space.

[0052] The local-global mass spectrum feature encoding module is used to perform feature encoding on mass spectrum data to obtain local and global spectrum feature embedded encoding (i.e., local-global spectrum feature embedded encoding).

[0053] In this embodiment, the local-global mass spectrum feature encoding module comprehensively considers the spatial relationship between different spectrum peaks in the mass spectrum data, and integrates spatial information and local information features. The local-global spectrum feature embedding encoding includes global intensity ranking ratio, global half-intensity ranking ratio, local intensity ranking ratio, local half-intensity ranking ratio, global relative intensity size, and local relative size, as follows: (1) Global strength ranking ratio ( ): is the ranking of each spectral peak intensity from large to small, Represents the total number of spectral peaks.

[0054] (2) Global half-intensity ranking ratio ( ): The corresponding ranking after the intensity of each spectral peak is halved and sorted from large to small, Represents the total number of spectral peaks.

[0055] (3) Local strength ranking ratio ( ): The intensity of the spectrum peak in the specified window size is defined as local information. In this embodiment, the window size is set to ,in is the ranking corresponding to the local peak intensity sorted from large to small, Represents the total number of local spectral peaks.

[0056] (4) Local half-intensity ranking ratio ( ): The corresponding ranking after the local peak intensity is halved and sorted from large to small, Represents the total number of local spectral peaks.

[0057] (5) Global relative strength ( ): is the size of a peak, It is the spectral peak with the highest intensity among all spectral peaks.

[0058] (6) Local relative size ( ): is the size of a peak, It is the spectral peak with the largest intensity among the local spectral peaks.

[0059] The local-global mass spectrum feature encoding module can construct the above six features into one The matrix is ​​denoted as , and then the matrix is ​​passed through a conversion network branch Mapped into a high-dimensional space, pattern encoding independent of the spectral peak order is achieved to obtain the final local-global spectral feature embedding encoding.

[0060] The mass spectrum peak fusion coding module is used to perform feature fusion on the mass spectrum peak embedding coding and the local and global spectrum feature embedding coding to obtain the mass spectrum peak fusion coding.

[0061] The fusion formula for each peak is as follows: ; in, , , represents the embedded coding result of the i-th peak; The mass-to-charge ratio represents the mass spectrum information; Represents the intensity of mass spectrum information; Represents the local-global spectral feature embedding encoding processed by the transformation network branch.

[0062] All encoders are used to generate mass spectrum peak fusion embedding features based on mass spectrum peak fusion encoding.

[0063] Specifically, in the calculation process of mass spectrum peak fusion embedding features, 6 standard Transformer architecture encoders are used. Each encoder consists of a multi-head self-attention module and a feedforward network to analyze the mass spectrum peak fusion coding and generate mass spectrum peak fusion embedding features.

[0064] The high-level semantic features extracted by the TSARseqNovo model can effectively help the subsequent semi-autoregressive decoding module and mask optimization decoding module to improve the feature extraction capabilities, so as to achieve the purpose of more effectively identifying and interpreting complex features in mass spectrometry data.

[0065] Among them, the basic operation of the encoder can be expressed as the following formula: ; in, , and They represent the three matrix representations learned from the linear mapping components in the standard Transformer architecture encoder. Therefore, the output of the encoder part can be recorded as: ; in, represents the encoders from the first to the kth, Code for mass spectrum peak fusion.

[0066] The peptide sequencing method provided in this embodiment adopts a new mass spectrum peak information fusion encoding module, proposes a new encoding strategy, integrates the mass spectrum peak encoding of mass spectrum data and the local-global mass spectrum feature encoding, so that the mass spectrum peak fusion embedded features finally obtained are not limited to mass-to-charge ratio and intensity information. This method can significantly enhance the performance of the model.

[0067] In some embodiments, an initial prediction is performed based on the mass spectrometry peak fusion embedded features to obtain an amino acid prediction result of the peptide to be sequenced, including: generating amino acid coding information of the peptide to be sequenced; and performing an initial prediction based on the amino acid coding information and the mass spectrometry peak fusion embedded features to obtain an amino acid prediction result of the peptide to be sequenced.

[0068] In the TSARseqNovo model, in order to ensure that the model can effectively consider the information of a single amino acid, this embodiment introduces a dynamic amino acid encoding method in the semi-autoregressive decoding module, which takes into account the mass, position and specific characteristics of each amino acid, so that when the model training is completed, the corresponding encoding can effectively capture the relationship between peptide segments and the characteristics of amino acids, such as their mass and relative position information.

[0069] Specifically, the semi-autoregressive decoding module can encode amino acid information and generate amino acid coding information of the peptide to be sequenced. The amino acid coding information includes the amino acid identifier of the peptide to be sequenced. The predicted sequence of the amino acid identifier uses [b] as the starting mark, which is used together with the mass spectrum peak fusion embedding feature as the input of the semi-autoregressive decoding module (also known as the SAR decoder) to calculate its hidden state vector and new amino acid identifier; then the predicted amino acid identifier is re-encoded, and new amino acid identifiers are generated repeatedly and iteratively until the stop mark [e] is encountered or the maximum length of 100 is reached. The amino acid encoding process of the semi-autoregressive decoding module can be expressed as the following formula: ; ; In the above formula, Represents the hidden unit vector of the last layer of the SAR decoder; Represents the sequence result predicted by the SAR decoder; Represents the information code corresponding to the previous amino acid output by the SAR decoder. It should be noted that when the length of the current peptide sequence reaches 100 or encounters the end symbol [e], it needs to be reset to [b].

[0070] The process repeats the prediction until a predetermined stop identifier [e] is encountered or the number of amino acids reaches 100. During this process, for each set of prediction results, the amino acid identifiers with a prediction confidence less than a preset threshold are screened out and marked as [m] to obtain the amino acid prediction results of the peptide to be sequenced.

[0071] It should be noted that in the semi-autoregressive decoding module, the input data is the mass spectrum peak fusion embedded features and the amino acid coding information, and the output amino acid prediction results include multiple amino identifiers of the peptide to be sequenced and the prediction confidence of each amino identifier. The semi-autoregressive decoding module introduces the idea of ​​semi-autoregression. The semi-autoregressive decoding module integrating this idea can process k amino coding information in parallel. Compared with the traditional method of the CasaNovo model that can only process one amino coding information at a time, the method of this embodiment can effectively improve the reasoning efficiency of the model.

[0072] See also Figures 3 to 5 , Figure 3is one of the schematic diagrams of the parallel reasoning of the semi-autoregressive decoding module provided by the present invention, Figure 4 This is the second schematic diagram of the parallel reasoning of the semi-autoregressive decoding module provided by the present invention. Figure 5 It is a schematic diagram of the processing flow of the semi-autoregressive decoding module provided by the present invention.

[0073] like Figure 3 and Figure 5 As shown in the figure, for a peptide segment with a length of M to be sequenced, it can be divided into groups are processed simultaneously; Figure 4 As shown, each group may include k amino acid encoding information.

[0074] Accordingly, in order to achieve parallel processing, the SAR decoder also improves its corresponding mask matrix. The specific formula is as follows: in, ,and Represents a floating point operation. This operation allows the attention mechanism to focus on up to k subsequent positions from a given mark, thereby achieving parallel processing. Its attention mechanism uses the following formula: ; in, , , Represent the Query matrix, Key matrix and Value matrix in the attention mechanism respectively; This modification of the traditional attention mechanism can support parallel group prediction functions and enhance the computational efficiency of the model for related peptide features.

[0075] Different from the step-by-step prediction strategy of the standard Transformer model adopted by the CasaNovo model, the peptide sequencing method provided in this embodiment introduces the idea of ​​semi-autoregression. While realizing the parallel prediction analysis of multiple amino groups, the historical amino group information of the model in the analysis of the last generated result is comprehensively considered in the analysis result, so that the final generated result can integrate more features, effectively improve the accuracy of the model prediction analysis results, and greatly improve the reasoning speed of the model. Compared with the baseline model, its reasoning efficiency is improved by 2-3 times.

[0076] In some embodiments, the amino acid prediction results include multiple amino acid identifiers of the peptide to be sequenced and the prediction confidence of each amino acid identifier; optimizing prediction is performed based on the amino acid prediction results to obtain peptide sequencing results, including: determining a target amino acid identifier based on the amino acid prediction results; the target amino acid identifier is an amino acid identifier with a prediction confidence less than a preset threshold; optimizing prediction is performed based on the target amino acid identifier and mass spectrum peak fusion embedding features to generate peptide sequencing results.

[0077] The masked optimized decoding module (also known as the MR decoder) and the semi-autoregressive decoding module both use a simplified version of the self-attention mechanism rather than the multi-head attention mechanism. The main function of the masked optimized decoding module is to determine the amino identifiers in the overall peptide segment whose prediction confidence is less than a preset threshold; and based on the amino identifiers with a prediction confidence less than a preset threshold and the mass spectrum peak fusion embedding features, predict again and regenerate amino identifiers with a prediction confidence greater than or equal to the preset threshold, until the prediction confidence of all the predicted amino identifiers is greater than or equal to the preset threshold, then generate the final peptide sequencing result, in order to improve the accuracy of the overall sequence.

[0078] It should be noted that in the initial stage of model training, the model's prediction results have great uncertainty, so its confidence score is set low. As the model is gradually optimized, the model prediction accuracy gradually improves. Therefore, during the training process, the confidence threshold can be gradually increased to ensure the reliability of the model results. In this process, amino identifiers with prediction confidence less than the preset threshold will also be marked as [m] for further analysis. The processing flow of the mask optimization decoding module can be expressed as the following formula: ; ; in, Represents the hidden state of the last layer of the MR decoder; are amino identifiers whose prediction confidence is less than a preset threshold; is the optimized peptide result generated by the MR decoder. The optimized output of the MR decoder is regarded as the final result prediction in the TSARseqNovo model.

[0079] The peptide sequencing method provided in this embodiment optimizes the mask structure of the existing CasaNovo model, introduces a secondary optimization module into the model, namely, a mask optimization decoding module, and performs secondary prediction on amino identifiers whose prediction confidence in the prediction results of the semi-autoregressive decoding module is less than a preset threshold, thereby achieving fine optimization and adjustment of the model prediction and improving the overall prediction accuracy of the model.

[0080] In some embodiments, the peptide sequencing model is trained using a preset loss function; wherein the preset loss function is determined based on a first loss function and a second loss function, the first loss function is a cross entropy loss function of a semi-autoregressive decoding module, and the second loss function is a cross entropy loss function of a mask optimized decoding module.

[0081] Specifically, the peptide sequencing model is trained using a preset loss function, which takes into account the cross entropy (CE) loss of the semi-autoregressive decoding module and the cross entropy loss of the mask optimization decoding module; among them, the semi-autoregressive decoding module calculates the loss between the predicted peptides and the true peptides, and the mask optimization decoding module focuses on calculating the loss between the optimized final peptides and the true peptides.

[0082] Preset loss function The calculation formula is as follows: ; in, is the cross entropy loss of the semi-autoregressive decoding module, is the first loss function; The cross entropy loss of the decoder module is optimized for the mask, is the second loss function.

[0083] In order to evaluate the performance of the TSARseqNovo model in this example, it was compared with other established peptide sequence analysis tools, such as the PEAKs model, the Novor model, the DeepNovo model, the PointNovo model, and the CasaNovo model. The focus was on indicators such as the prediction accuracy at the peptide level, the recall rate of amino acid labels, and the amino acid accuracy, which were consistent with the evaluation criteria previously applied to the DeepNovo model, the PointNovo model, and the CasaNovo model.

[0084] Two main datasets were used in the model evaluation: the nine-species benchmark dataset and the merged high-resolution dataset. The nine-species benchmark dataset was used for the evaluation of the DeepNovo model, the PointNovo model, and the CasaNovo model. The dataset contains about 1.5 million MS spectra from nine different experiments. For the true value in the dataset, the peptide information with a false discovery rate of less than or equal to 1% in the papers related to the CasaNovo model was used as the true value for model training and evaluation. The merged high-resolution dataset contains about 700,000 MS spectra from seven high-resolution proteomics datasets PXD008808, PXD011246, PXD012645, PXD012979, PXD008844, and PXD010559, and its true value comes from the π-HelixNovo study.

[0085] In terms of validation methods, the leave-one-out cross-validation method previously used by the DeepNovo model and CasaNovo model was adopted. This method uses eight of the nine species datasets for training and retains the ninth for testing. For the merged dataset, the strategy of the PointNovo model was borrowed, using PXD008808, PXD011246, PXD012645, and PXD012979 as training sets. The remaining datasets ABRF, PXD008844, and PXD010559 were used for testing, thereby comprehensively evaluating the performance of the model under different MS conditions.

[0086] See also Figures 6 to 8 , Figure 6 This is one of the schematic diagrams of the evaluation results of different models provided by the present invention on nine species data sets. Figure 7 This is the second schematic diagram of the evaluation results of different models provided by the present invention on nine species data sets. Figure 8 This is the third schematic diagram of the evaluation results of different models provided by the present invention on nine species data sets.

[0087] like Figures 6 to 8 As shown in Figure 3, the TSARseqNovo model outperforms many existing models.

[0088] Depend on Figure 6 It can be seen that compared with the traditional PEAKs model and Novor model, the comprehensive performance of the TSARseqNovo model is 18.0% higher than that of the PEAKs model and 34.8% higher than that of the Novor model. Compared with the existing DeepNovo model, PointNovo model and CasaNovo model based on deep learning, the TSARseqNovo model is 12.6% higher than the DeepNovo model in terms of peptide level accuracy, 11.4% higher than the PointNovo model, and 11.4%4.9% higher than the CasaNovo model. There are 1.9%, 3.9%, 4.1%, 5.1%, 8.3%, 3.8% and 6.8% improvements in Human, Yeast, M.mazei, Honeybee, Tomato, Bacilus and Clam bacteria respectively.

[0089] Depend on Figure 7 It can be seen that in terms of amino acid accuracy, the TSARseqNovo model of this embodiment has improvements of 6.1%, 23.4%, 5.1%, 3.7% and 4.7% compared with the PEAKs model, Novor model, DeepNovo model, PointNovo model and CasaNovo model, respectively.

[0090] Depend on Figure 8 It can be seen that in terms of recall rate, the TSARseqNovo model of this embodiment has improvements of 6.8%, 23.8%, 5.5%, 3.8% and 5.0% compared with the PEAKs model, the Novor model, the DeepNovo model, the PointNovo model and the CasaNovo model.

[0091] See also Figures 9 to 11 , Fig. 9 This is one of the schematic diagrams of the evaluation results of different models provided by the present invention on the combined high-resolution data set. Fig.10 This is the second schematic diagram of the evaluation results of different models provided by the present invention on the combined high-resolution data set. Fig.11 This is the third schematic diagram of the evaluation results of different models provided by the present invention on the merged high-resolution data set.

[0092] In the merged high-resolution dataset, the TSARseqNovo model of this embodiment is trained on the sub-datasets PXD008808, PXD011246, PXD012645, and PXD012979, and is inferred and verified on the sub-datasets ABRF, PXD008844, and PXD010559.

[0093] Depend on Fig. 9 It can be seen that the performance of the TSARseqNovo model significantly exceeds that of the CasaNovo model and the PointNovo model, achieving performance improvements of 7.0% and 3.3% respectively.

[0094] However, the TSARseqNovo model showed some decrease in the prediction accuracy and recall of amino acids.

[0095] Depend on Fig.10 and Fig.11 It can be seen that the TSARseqNovo model performed well on the ABRF dataset, with a 3.2% improvement in precision and a 3.6% improvement in recall compared to the PointNovo model. In contrast, the precision of the TSARseqNovo model on the PXD008844 dataset decreased by 1.6% and the recall decreased by 2.2%. On the PXD010559 dataset, the precision of the TSARseqNovo model on the PXD008844 dataset decreased by 1.6% and the recall decreased by 2.2%. The precision and recall of the PXD01 decreased by 3.8% and 3.1% respectively.

[0096] To further explore the performance of the TSARseqNovo model in terms of inference speed and accuracy, the Yeast dataset is used for further evaluation. The results are shown in Table 1.

[0097] Table 1

[0098] As shown in Table 1, the TSARseqNovo model explored the effect of the parallel decoding length of decoding two amino acids and three amino acids on the inference speed and accuracy. The results show that increasing the decoding length of the TSARseqNovo model can significantly improve the inference speed of the model. Among them, when the inference length is set to 3, the processing speed can reach 234 mass spectrometry data per second, which is twice that of the CasaNovo model. At this speed, the model accuracy of the TSARseqNovo model (52.1%) exceeds the model accuracy of the CasaNovo model (50.5%). Although the accuracy of the TSARseqNovo model is slightly lower than that of the PointNovo model, its inference speed is about 10 times that of the PointNovo model. In addition, when the number of parallels is set to 3, the amino acid accuracy predicted by the TSARseqNovo model drops from 76.1% to 74.8%, the amino acid recall rate drops from 75.3% to 74.6%, and the peptide accuracy drops from 54.4% to 52.1%. This shows that although a longer decoding length improves the inference speed, it will directly affect the accuracy of the TSARseqNovo model.

[0099] Overall, the TSARseqNovo model has a significant improvement in inference speed while still maintaining high accuracy, and therefore has great application significance as a powerful tool for peptide sequence analysis applications.

[0100] The present invention also provides a peptide sequencing device. Fig.12 , Fig.12 12 is a schematic diagram of the structure of the peptide sequencing device provided by the present invention. In this embodiment, the peptide sequencing device includes an acquisition module 1210 and a sequencing module 1220.

[0101] The acquisition module 1210 is used to acquire mass spectrum data of the peptide segment to be sequenced.

[0102] The sequencing module 1220 is used to input the mass spectrometry data into a peptide sequencing model to obtain a peptide sequencing result output by the peptide sequencing model.

[0103] The peptide sequencing model is obtained by training an initial model constructed based on the TSARseqNovo algorithm based on sample mass spectrometry data and sample peptide sequencing results corresponding to the sample mass spectrometry data.

[0104] In some embodiments, the peptide sequencing model includes a mass spectrum peak information fusion encoding module, a semi-autoregressive decoding module and a mask optimization decoding module; the mass spectrum peak information fusion encoding module is used to extract features from mass spectrum data and obtain mass spectrum peak fusion embedded features; the semi-autoregressive decoding module is used to perform initial predictions based on the mass spectrum peak fusion embedded features to obtain amino acid prediction results of the peptide to be sequenced; the mask optimization decoding module is used to perform optimized predictions based on the amino acid prediction results to obtain peptide sequencing results.

[0105] In some embodiments, the mass spectrum peak information fusion coding module includes a mass spectrum peak coding module, a local and global mass spectrum feature coding module, a mass spectrum peak fusion coding module and multiple encoders; the mass spectrum peak coding module is used to feature encode the mass spectrum data to obtain the mass spectrum peak embedded coding; the local and global mass spectrum feature coding module is used to feature encode the mass spectrum data to obtain the local and global spectrum feature embedded coding; the mass spectrum peak fusion coding module is used to feature fuse the mass spectrum peak embedded coding and the local and global spectrum feature embedded coding to obtain the mass spectrum peak fusion coding; all encoders are used to generate mass spectrum peak fusion embedded features based on the mass spectrum peak fusion coding.

[0106] In some embodiments, an initial prediction is performed based on the mass spectrometry peak fusion embedded features to obtain an amino acid prediction result of the peptide to be sequenced, including: generating amino acid coding information of the peptide to be sequenced; and performing an initial prediction based on the amino acid coding information and the mass spectrometry peak fusion embedded features to obtain an amino acid prediction result of the peptide to be sequenced.

[0107] In some embodiments, the amino acid prediction results include multiple amino acid identifiers of the peptide to be sequenced and the prediction confidence of each amino acid identifier; optimizing prediction is performed based on the amino acid prediction results to obtain peptide sequencing results, including: determining a target amino acid identifier based on the amino acid prediction results; the target amino acid identifier is an amino acid identifier with a prediction confidence less than a preset threshold; optimizing prediction is performed based on the target amino acid identifier and mass spectrum peak fusion embedding features to generate peptide sequencing results.

[0108] In some embodiments, the peptide sequencing model is trained using a preset loss function; wherein the preset loss function is determined based on a first loss function and a second loss function, the first loss function is a cross entropy loss function of a semi-autoregressive decoding module, and the second loss function is a cross entropy loss function of a mask optimized decoding module.

[0109] The invention also provides an electronic device. Fig.13 is a schematic diagram of the structure of the electronic device provided by the present invention, such as Fig.13As shown, the electronic device may include: a processor 1310, a communications interface 1320, a memory 1330, and a communications bus 1340, wherein the processor 1310, the communications interface 1320, and the memory 1330 communicate with each other through the communications bus 1340. The processor 1310 may call the logic instructions in the memory 1330 to execute the peptide sequencing method.

[0110] In addition, the logic instructions in the above-mentioned memory 1330 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0111] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the computer program is executed by a processor to execute the peptide sequencing method provided by the above methods.

[0112] The present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the peptide sequencing methods provided by the above methods.

[0113] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0114] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0115] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A peptide sequencing method, characterized in that: include: Obtaining mass spectrometry data of peptides to be sequenced; Inputting the mass spectrometry data into a peptide sequencing model to obtain a peptide sequencing result output by the peptide sequencing model; The peptide sequencing model is obtained by training an initial model constructed based on the TSARseqNovo algorithm based on sample mass spectrometry data and sample peptide sequencing results corresponding to the sample mass spectrometry data.

2. The peptide sequencing method according to claim 1, characterized in that: The peptide sequencing model includes a mass spectrum peak information fusion encoding module, a semi-autoregressive decoding module and a mask optimization decoding module; The mass spectrum peak information fusion encoding module is used to extract features from the mass spectrum data to obtain mass spectrum peak fusion embedded features; The semi-autoregressive decoding module is used to make an initial prediction based on the mass spectrum peak fusion embedding feature to obtain the amino acid prediction result of the peptide segment to be sequenced; The mask optimization decoding module is used to perform optimization prediction based on the amino acid prediction result to obtain the peptide segment sequencing result.

3. The peptide sequencing method according to claim 2, characterized in that: The mass spectrum peak information fusion coding module includes a mass spectrum peak coding module, a local and global mass spectrum feature coding module, a mass spectrum peak fusion coding module and a plurality of encoders; The mass spectrum peak encoding module is used to perform feature encoding on the mass spectrum data to obtain mass spectrum peak embedded code; The local and global mass spectrum feature encoding module is used to perform feature encoding on the mass spectrum data to obtain local and global spectrum feature embedded encoding; The mass spectrum peak fusion coding module is used to perform feature fusion on the mass spectrum peak embedded coding and the local and global spectrum feature embedded coding to obtain the mass spectrum peak fusion coding; All of the encoders are used to generate the mass spectrum peak fusion embedding features based on the mass spectrum peak fusion encoding.

4. The peptide sequencing method according to claim 2, characterized in that: The initial prediction based on the mass spectrum peak fusion embedding feature to obtain the amino acid prediction result of the peptide segment to be sequenced includes: Generating amino coding information of the peptide to be sequenced; An initial prediction is performed based on the amino acid encoding information and the mass spectrum peak fusion embedding feature to obtain an amino acid prediction result of the peptide to be sequenced.

5. The peptide sequencing method according to claim 2, characterized in that: The amino acid prediction result includes a plurality of amino acid identifiers of the peptide segment to be sequenced and a prediction confidence of each of the amino acid identifiers; The optimizing prediction based on the amino acid prediction result to obtain the peptide sequencing result includes: Based on the amino acid prediction result, determining a target amino group identifier; the target amino group identifier is an amino group identifier whose prediction confidence is less than a preset threshold; An optimized prediction is performed based on the target amino identifier and the mass spectrum peak fusion embedding feature to generate the peptide sequencing result.

6. The peptide sequencing method according to claim 2, characterized in that: The peptide sequencing model is trained using a preset loss function; Among them, the preset loss function is determined based on a first loss function and a second loss function, the first loss function is the cross entropy loss function of the semi-autoregressive decoding module, and the second loss function is the cross entropy loss function of the mask optimization decoding module.

7. A peptide sequencing device, characterized in that: include: An acquisition module is used to obtain mass spectrometry data of peptides to be sequenced; A sequencing module, used to input the mass spectrometry data into a peptide sequencing model to obtain peptide sequencing results output by the peptide sequencing model; The peptide sequencing model is obtained by training an initial model constructed based on the TSARseqNovo algorithm based on sample mass spectrometry data and sample peptide sequencing results corresponding to the sample mass spectrometry data.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the peptide sequencing method according to any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the peptide sequencing method according to any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the peptide sequencing method according to any one of claims 1 to 6 is implemented.

Images