A method and system for analyzing peptide spectrum matching data of small molecule peptide donkey-hide gelatin

Abstract

The invention relates to the technical field of protein data analysis, and discloses a polypeptide spectrum matching data analysis method and system for a small molecule peptide donkey-hide gelatin. The method comprises the following steps: acquiring historical protein identification data and a theoretical mass spectrum protein database; constructing a strategy network model based on the historical protein data, wherein the input of the strategy network model is a comprehensive feature vector, and the output is an action in a current state; executing the action output by the strategy network model, applying the action to a new protein, obtaining new polypeptide spectrum data, and obtaining corresponding protein identification results according to the obtained new polypeptide spectrum data analysis; the analysis method of the invention greatly improves the efficient and accurate identification of protein types in complex samples by integrating historical data, constructing an intelligent decision-making model, and using a strategy network for dynamic adjustment.

Description

Technical Field

[0001] The present invention relates to the field of protein data analysis, and more specifically, to a method and system for analyzing polypeptide spectrum matching data of small molecule peptide donkey-hide gelatin. Background Art

[0002] The main component of donkey-hide gelatin is protein (about 80%), but the digestion and absorption of protein requires the participation of proteases in degradation. With the development of modern bioengineering technology, the biomimetic enzymatic hydrolysis method can be used to obtain donkey-hide gelatin hydrolysates, namely donkey-hide gelatin protein peptides. In order to ensure the quality and purity of peptide donkey-hide gelatin products, it is necessary to identify the specific protein components contained in donkey-hide gelatin protein peptides. In order to accurately identify all types of proteins in donkey-hide gelatin protein peptides, it is necessary to separate different types of proteins into individual ones before identification, and then identify the separated multiple proteins separately.

[0003] Currently, the common method of protein identification is to identify peptides and proteins in peptide spectrum data by comparing them with a known protein database. However, traditional protein identification methods usually use deterministic rules or scoring functions. This use of a single fixed rule or scoring function often cannot fully adapt to all situations and is difficult to adapt to the complex and changeable characteristics of mass spectrometry data, resulting in low accuracy in identifying proteins corresponding to peptide spectrum data. In addition, the detection methods for protein identification are prone to repetition and are not the optimal detection method. Summary of the invention

[0004] The present invention provides a method and system for analyzing polypeptide spectrum matching data of a small molecule peptide donkey-hide gelatin, which solves the technical problems raised in the above-mentioned background technology.

[0005] The present invention provides a method for analyzing polypeptide spectrum matching data of a small molecule peptide donkey-hide gelatin, comprising the following steps:

[0006] Step 101, obtaining historical protein identification data and a theoretical mass spectrum protein database, wherein the historical protein identification data includes peptide spectrum data and enzymatic hydrolysis information, wherein the peptide spectrum data includes the mass-to-charge ratio value and abundance of peptide segments and the mass-to-charge ratio value and abundance of b / y ions in fragmentation mode, and the enzymatic hydrolysis information includes the type of enzyme used, and each enzyme corresponds to a specific enzyme cleavage site, enzymatic hydrolysis time and conditions;

[0007] Step 102, constructing a policy network model based on historical protein data, the input of the policy network model is a comprehensive feature vector, and the output is an action in the current state;

[0008] Step 103, execute the action output by the strategy network model, apply the action to the new protein, obtain new peptide spectrum data, and obtain the corresponding protein identification result according to the obtained new peptide spectrum data analysis.

[0009] In a preferred embodiment, in step 101, the historical protein identification data is denoised to remove noise and low-quality peaks; the mass-to-charge ratio and abundance are normalized to ensure that the data are on a uniform scale; the peaks of the experimental mass spectrometry data and the theoretical mass spectrometry data are aligned to ensure that the peaks of the same peptide in different spectra can match; the historical peptide spectrum data is extracted, and the mass-to-charge ratio, abundance and type of fragmentation products of each peptide and its fragmentation product are used as features.

[0010] In a preferred embodiment, the policy network model includes:

[0011] Definition of state space:

[0012] Enzymatic hydrolysis information: type of enzyme used;

[0013] Peptide spectrum data: mass-to-charge ratio and abundance of peptide segments, mass-to-charge ratio and abundance of b / y ions in fragmentation mode;

[0014] Identified peptide sequence: confirmed peptide sequence and its location in the protein;

[0015] Database query status: the number of peptides matching the known database, the degree of matching, and the coverage of the matching peptides;

[0016] Resource consumption: elapsed time, reagent consumption;

[0017] The enzymatic hydrolysis information, peptide spectrum data, identified peptide sequences, database query status and resource consumption information are encoded and concatenated into a comprehensive feature vector.

[0018] In a preferred embodiment, the policy network model includes: definition of action space:

[0019] Select an enzymatic strategy: Choose a specific enzyme;

[0020] Mass spectrometry parameter adjustment: adjust scanning range, resolution, fragmentation mode, and fragmentation energy;

[0021] Database query: select the database to query and set the search parameters;

[0022] Experimental process control: decide whether to continue enzymatic hydrolysis, repeat mass spectrometry analysis, or change the direction of the experiment.

[0023] In a preferred embodiment, the strategy network model includes: establishing a reward function for the output action of the strategy network model, the reward goal is to select the optimal enzymatic hydrolysis strategy, mass spectrometry parameter setting and database query method to maximize the accuracy and efficiency of identification, and the reward function is as follows:

[0024] Rewards given when the identified protein exactly matches the real protein:

[0025] R1=+100

[0026] Rewards given when the identified protein partially matches the real protein:

[0027]

[0028] Among them, Coverage current Indicates the coverage of the currently identified peptide. max Indicates the maximum coverage;

[0029] The reward for each 1% increase in peptide coverage is:

[0030] R3=+0.5

[0031] The calculation formula for resource efficiency bonus is:

[0032]

[0033] Among them, Resource preset Indicates the preset resource threshold, Resource used Indicates the actual consumption of resources;

[0034] The rewards deducted for incorrect identification are:

[0035] R5=-50

[0036] For every 10% of excess over the resource budget, the reward deducted is:

[0037] R6=-10

[0038] The rewards deducted for repeated invalid operations are:

[0039] R7=-1

[0040] The rewards for exploration are:

[0041] R8 = +2

[0042] The rewards for continued learning are:

[0043] R9=+1

[0044] The total reward is calculated as follows:

[0045] R total

[0046] =R1·I1+R2·I2+R3·I3+R4·I4-R5·I5-R6·I6-R7·I7+R8·I8+R9·I9

[0047]

[0048] Among them, R total represents the total reward, R1 represents the reward value obtained when the protein completely matches the real protein, R2 represents the reward value obtained when the protein partially matches the real protein, R3 represents the reward value obtained for each 1% increase in peptide coverage, R4 represents the reward value obtained when the resources are below a certain proportion of the preset threshold to complete the identification, R5 represents the penalty value obtained when the identified protein does not match the real one, R6 represents the penalty value obtained when the preset resource threshold is exceeded, R7 represents the penalty value obtained when invalid operations are repeated, R8 represents the reward value obtained by exploration, and R9 represents the reward value obtained by continuous learning. Coverage current Indicates the coverage of the currently identified peptide. max Indicates the theoretical maximum coverage, Resource preset Indicates the preset resource threshold, Resource used represents the actual consumption of resources, and I1, I2, I3, I4, I5, I6, I7, I8, and I9 represent the 1st, 2nd, 3rd, 4th, 5th, 6th, 7th, 8th, and 9th condition coefficients respectively.

[0049] In a preferred embodiment, the policy network model comprises: an input layer, a first hidden layer, a second hidden layer and an output layer;

[0050] The comprehensive feature vector is input into the input layer, the input layer outputs the comprehensive feature vector to the first hidden layer to generate a first intermediate feature vector, the first hidden layer outputs the first intermediate feature vector to the second hidden layer to generate a second intermediate feature vector, the second hidden layer outputs the second intermediate feature vector to the output layer to generate an action vector, and the action vector mapping obtains the probability of performing the corresponding action in the current state.

[0051] In a preferred embodiment, the calculation formula of the first hidden layer is:

[0052] Z1=W1x+b1

[0053] H1=f(Z1)

[0054] Among them, H1 represents the first intermediate feature vector, f represents the ReLU activation function, Z1 represents the result of the linear transformation of the comprehensive feature vector, x represents the comprehensive feature vector, W1 represents the first weight matrix, and b1 represents the first bias term;

[0055] The calculation formula for the second hidden layer is:

[0056] Z2=W2H1+b2

[0057] H2=f(Z2)

[0058] Wherein, H2 represents the second intermediate eigenvector, f represents the ReLU activation function, Z2 represents the result after the linear transformation of the first intermediate eigenvector, H1 represents the first intermediate eigenvector, W2 represents the second weight matrix, and b2 represents the second bias term;

[0059] The calculation formula of the output layer is:

[0060] Z3=W3H2+b3

[0061]

[0062] Among them, Z3 represents the action feature, W3 represents the third weight matrix, H2 represents the second intermediate feature vector, b3 represents the third bias term, and y i represents the probability of executing the i-th action in the current state, represents the i-th component in the action feature vector, represents the jth component in the action feature vector, and i, j∈n, n represents the number of all executable actions, and e represents a natural constant.

[0063] The policy network model inputs the comprehensive feature vector for the hth time, and then outputs the action of mass spectrometry analysis of the hydrolyzed protein for the h+1th time;

[0064] The action of the first hydrolysis protein mass spectrometry analysis comes from manual experience or historical parameters used in similar protein mass spectrometry analysis. When h = 0, the comprehensive feature vector comes from the data after the first hydrolysis protein mass spectrometry analysis;

[0065] The policy network model continuously iterates and generates actions according to the above method. When any termination condition is reached, the step of generating actions is stopped. The termination conditions include:

[0066] The number of generated actions reaches the set upper limit; for example, 20 times;

[0067] The peptide coverage reaches the set value; for example, 96%. Generally speaking, this value needs to be set to be greater than 85%.

[0068] In a preferred embodiment, the loss function of the policy network model is the Clip Loss in PPO or the policy loss function of A3C.

[0069] A peptide spectrum matching data analysis system for small molecule peptide donkey-hide gelatin, comprising the following modules:

[0070] Data acquisition module, used to obtain historical protein identification data and theoretical mass spectrometry protein database;

[0071] The process optimization module is used to build a strategy network model and output the best enzymatic hydrolysis strategy and mass spectrometry parameter configuration that should be adopted under the current state;

[0072] The peptide spectrum data analysis and identification module is used to obtain the corresponding protein identification results based on the new peptide spectrum data analysis.

[0073] A storage medium stores non-transitory computer-readable instructions. When the non-transitory computer-readable instructions are executed by a computer, the steps in the above-mentioned method for analyzing polypeptide spectrum matching data of a small molecule peptide donkey-hide gelatin can be executed.

[0074] The beneficial effects of the present invention are as follows: the analysis method of the present invention provides a new method for efficient and accurate identification of proteins in complex samples by integrating historical data, constructing an intelligent decision-making model, and using a strategy network for dynamic adjustment. This not only improves the scientific nature of the identification, but also accelerates the research process, especially when dealing with samples containing multiple complex components such as donkey-hide gelatin. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Figure 1 It is a flow chart of the polypeptide spectrum matching data analysis method of the small molecule peptide donkey-hide gelatin of the present invention.

[0076] Figure 2 It is a block diagram of the polypeptide spectrum matching data analysis system of the small molecule peptide donkey-hide gelatin of the present invention.

[0077] In the figure: 100, data acquisition module; 200, process optimization module; 300, peptide spectrum data analysis and identification module. DETAILED DESCRIPTION

[0078] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that the discussion of these embodiments is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the contents of this specification. Each example may omit, replace or add various processes or components as needed. In addition, the features described relative to some examples may also be combined in other examples.

[0079] like Figure 1 As shown, a method for analyzing peptide spectrum matching data of a small molecule peptide donkey-hide gelatin comprises the following steps:

[0080] Step 101, obtaining historical protein identification data and a theoretical mass spectrum protein database, wherein the historical protein identification data includes peptide spectrum data and enzymatic hydrolysis information, wherein the peptide spectrum data includes the mass-to-charge ratio value and abundance of peptide segments and the mass-to-charge ratio value and abundance of b / y ions in fragmentation mode, and the enzymatic hydrolysis information includes the type of enzyme used, and each enzyme corresponds to a specific enzyme cleavage site, enzymatic hydrolysis time and conditions;

[0081] It should be noted that the theoretical mass spectrometry protein database constructed based on known proteins includes multiple database subsets. The protein database can be classified according to different standards and characteristics to form multiple database subsets. For example, different subsets are formed according to the three-dimensional structure and resolution of proteins, such as high-resolution structure subsets, membrane protein subsets, etc.; proteins are classified according to common types to form multiple database subsets, such as general databases, specific species databases, customized databases, etc.

[0082] De-noising the historical protein identification data to remove noise and low-quality peaks; normalizing the mass-to-charge ratio and abundance to ensure that the data are on a uniform scale;

[0083] Peak alignment of experimental mass spectrometry data and theoretical mass spectrometry data was performed to ensure that the peaks of the same peptide in different spectra could match;

[0084] Extract historical peptide spectrum data, and use the mass-to-charge ratio, abundance, and type of fragmentation products of each peptide segment and its fragmentation products as features;

[0085] The characteristics of each peptide and its fragmentation products are converted into vector form;

[0086] A vector representation of a peptide can include its mass-to-charge ratio, abundance, and fragmentation pattern information (which can infer the composition and structure of the protein and help identify the identity and properties of unknown proteins);

[0087] The vector representations of peptides and their fragmentation products are arranged in order to form a sequence; each sequence represents the complete information of a peptide and its fragmentation products;

[0088] If the number of peptides is inconsistent, padding (such as zero padding) can be used to unify the length;

[0089] Combine the sequence representations of multiple peptides and their fragmentation products into a multidimensional tensor; this tensor serves as the input of the neural network;

[0090] For example, the feature vector of a peptide is represented as:

[0091] c X =(mz1, FD1, SL1, mz2, FD2, SL2,…, mzn , F.D. n , SL n )

[0092] Among them, mz1, mz2, mz n FD1, FD2, FD n SL1, SL2, SL n They represent the fragmentation patterns of the 1st, 2nd and nth peaks in the mass spectrum of the peptide, respectively;

[0093] Combine feature vectors of multiple peptides into a single sequence:

[0094]

[0095] in, They represent the feature vectors of the 1st, 2nd, and mth peptide segments respectively;

[0096] Construct the input peptide spectrum feature tensor:

[0097]

[0098] in, denote the 1st, 2nd, and kth sequences respectively;

[0099] Similarly, the protein information in the theoretical mass spectrometry protein database is converted into vector and sequence representation; the representation of each protein can include its amino acid sequence, theoretical mass spectrometry peak value and other information;

[0100] Step 102, constructing a policy network model based on historical protein data, the input of the policy network model is a comprehensive feature vector, and the output is an action in the current state;

[0101] It should be noted that the optimization goal of the strategy network model is to find the best path from a large number of possible operations through enzymatic hydrolysis, mass spectrometry analysis and database comparison, so as to efficiently and accurately identify the types of proteins in donkey-hide gelatin peptides; the task of the intelligent agent is to maximize the accuracy and efficiency of identification by selecting the optimal enzymatic hydrolysis strategy, mass spectrometry parameter settings and database query methods within a limited number of experimental steps.

[0102] The policy network model includes:

[0103] 1. Definition of state space:

[0104] Enzymatic hydrolysis information: type of enzyme used;

[0105] Enzyme hydrolysis information encoding: Use one-hot encoding to indicate the type of enzyme;

[0106] Peptide spectrum data: mass-to-charge ratio and abundance of peptide segments, mass-to-charge ratio and abundance of b / y ions in fragmentation mode;

[0107] Peptide spectrum data encoding: The m / z value and abundance of the peptides are converted into numerical vectors, which may be normalized to maintain the importance balance of each feature. The mass-to-charge ratio and abundance of the fragmentation products are also converted into numerical arrays;

[0108] Identified peptide sequence: confirmed peptide sequence and its location in the protein;

[0109] Database query status: the number of peptides matching the known database, the degree of matching (e.g., e-value, score), and the coverage of the matching peptides;

[0110] Sequence matching information encoding: Use binary encoding to indicate whether each expected peptide is successfully matched, as well as a quantitative indicator of the degree of matching;

[0111] Resource consumption: elapsed time, reagent consumption, etc., reflecting the economy and efficiency of current decision-making.

[0112] Resource consumption coding: expressing resource consumption in the form of remaining time and / or cost;

[0113] Integrated coding: The above information is serially integrated into a vector of fixed length to generate a comprehensive feature vector;

[0114] 2. Definition of action space:

[0115] Select an enzymatic strategy: Choose a specific enzyme;

[0116] Mass spectrometry parameter adjustment: adjust scanning range, resolution, fragmentation mode (CID, HCD, ETD, etc.), fragmentation energy;

[0117] Database query: select the database to be queried (general database, species-specific database, customized database), set search parameters (such as allowed modification types, error tolerance);

[0118] Experimental process control: decide whether to continue enzymatic digestion (when the initial enzymatic digestion conditions fail to produce enough or diverse peptides, resulting in low coverage of protein identification, the agent will choose to continue enzymatic digestion), repeat mass spectrometry analysis (when the initial mass spectrometry analysis results in low data quality due to insufficient sample preparation, improper instrument settings or random noise, and cannot obtain a clear peptide map or match to the protein database), change the experimental direction (when the existing experimental path, such as a specific combination of enzymatic digestion and mass spectrometry parameters, fails to effectively advance the protein identification process, it may be due to the special nature or complexity of the sample, the agent can decide to change the experimental direction);

[0119] 3. Establish reward function:

[0120] The goal of the reward function design is to motivate the agent to pursue efficient, accurate, and resource-economical decisions in the protein identification process;

[0121] Basic reward design:

[0122] 1. Rewards for correct identification:

[0123] If the identified protein completely matches the real protein, a high one-time reward will be given:

[0124] R1=+100

[0125] If the identified protein partially matches the real protein (i.e., some of the identified peptides belong to the target protein), rewards will be given according to the coverage of the matched peptides:

[0126]

[0127] Among them, Coverage current Indicates the coverage of the currently identified peptide. max Indicates the maximum coverage;

[0128] 2. Coverage Rewards:

[0129] For every 1% increase in peptide coverage (relative to the theoretical coverage of known protein sequences), rewards:

[0130] R3=+0.5

[0131] 3. Resource efficiency rewards:

[0132] If the resources (such as time and reagent consumption) are lower than a certain percentage of the preset threshold, the identification will be completed and rewards will be given according to the savings ratio:

[0133]

[0134] Among them, Resource preset Indicates the preset resource threshold, Resource used Indicates the actual consumption of resources;

[0135] 1. Penalty items:

[0136] Penalty for misidentification (identified protein does not match the real one):

[0137] R5=-50

[0138] If the resource budget (time or cost) is exceeded, a penalty will be deducted for every 10% overrun:

[0139] R6=-10

[0140] Repeating invalid operations (such as continuously selecting the same and invalid enzyme digestion conditions) will result in a penalty of:

[0141] R7=-1

[0142] Dynamically adjust rewards:

[0143] Exploration rewards: In the early stages of learning, to encourage exploration, rewards for trying new actions (such as using a new enzyme, changing

[0144] Experimental process) Give small rewards:

[0145] R8 = +2

[0146] Continuous learning rewards: If there is no significant progress for a period of time, provide a mild continuous learning reward:

[0147] R9=+1

[0148] The total reward is calculated as follows:

[0149] R total

[0150] =R1·I1+R2·I2+R3·I3+R4·I4-R5·I5-R6·I6-R7·I7+R8·I8+R9·I9

[0151]

[0152]

[0153] Among them, R total represents the total reward, R1 represents the reward value obtained when the protein completely matches the real protein, R2 represents the reward value obtained when the protein partially matches the real protein, R3 represents the reward value obtained for each 1% increase in peptide coverage, R4 represents the reward value obtained when the resources are below a certain proportion of the preset threshold to complete the identification, R5 represents the penalty value obtained when the identified protein does not match the real one, R6 represents the penalty value obtained when the preset resource threshold is exceeded, R7 represents the penalty value obtained when invalid operations are repeated, R8 represents the reward value obtained by exploration, and R9 represents the reward value obtained by continuous learning. Coverage current Indicates the coverage of the currently identified peptide. max Indicates the theoretical maximum coverage, Resource preset Indicates the preset resource threshold, Resource used represents the actual consumption of resources, I1, I2, I3, I4, I5, I6, I7, I8, and I9 represent the 1st, 2nd, 3rd, 4th, 5th, 6th, 7th, 8th, and 9th condition coefficients respectively;

[0154] The steps for storing experience are as follows:

[0155] Step 1: Design an experience unit to record the key information of each decision cycle (x, a, r, x'):

[0156] Among them, x represents the current feature representation of the protein, a represents the selected experimental condition or treatment method, r represents the reward value of the current state calculated according to the reward function, and x' represents the feature change of the protein after treatment;

[0157] Step 2: After each decision cycle iteration, observe and record the current state x and the action a taken; calculate and obtain the reward value r of this operation; update and record the new state of the next round;

[0158] Step 3: Build an experience replay pool to save experience units;

[0159] It should be noted that the capacity of the experience replay pool is designed according to the actual situation to ensure that the experience replay pool is large enough to accommodate diverse experience samples, while controlling its size to avoid excessive consumption of resources;

[0160] Each time a new experience is added, if the pool is full, the oldest experience is removed to ensure that the latest experience can be added;

[0161] Step 4: Regularly randomly extract samples from the experience replay pool for training and updating the policy network to promote learning stability and efficiency;

[0162] This maintains the dynamic balance of the experience replay pool and ensures that it contains the latest and representative interaction records.

[0163] The above storage experience can break the correlation between data and improve the stability of training.

[0164] Strategy network model: including input layer, first hidden layer, second hidden layer and output layer;

[0165] The comprehensive feature vector is input into the input layer, the input layer outputs the comprehensive feature vector to the first hidden layer to generate a first intermediate feature vector, the first hidden layer outputs the first intermediate feature vector to the second hidden layer to generate a second intermediate feature vector, the second hidden layer outputs the second intermediate feature vector to the output layer to generate an action vector, and the action vector mapping obtains the probability of executing the corresponding action in the current state;

[0166] The calculation formula for the first hidden layer is:

[0167] Z1=W1x+b1

[0168] H1=f(Z1)

[0169] Among them, H1 represents the first intermediate feature vector, f represents the ReLU activation function, Z1 represents the result of the linear transformation of the comprehensive feature vector, x represents the comprehensive feature vector, W1 represents the first weight matrix, and b1 represents the first bias term;

[0170] The calculation formula for the second hidden layer is:

[0171] Z2=W2H1+b2

[0172] H2=f(Z2)

[0173] Wherein, H2 represents the second intermediate eigenvector, f represents the ReLU activation function, Z2 represents the result after the linear transformation of the first intermediate eigenvector, H1 represents the first intermediate eigenvector, W2 represents the second weight matrix, and b2 represents the second bias term;

[0174] The calculation formula of the output layer is:

[0175] Z3=W3H2+b3

[0176]

[0177] Among them, Z3 represents the action feature, W3 represents the third weight matrix, H2 represents the second intermediate feature vector, b3 represents the third bias term, and y i represents the probability of executing the i-th action in the current state, represents the i-th component in the action feature vector, represents the jth component in the action feature vector, and i, j∈n, n represents the number of all executable actions, and e represents a natural constant;

[0178] Use general reinforcement learning training methods to train the policy network model, or configure the loss function to be the Clip Loss in PPO or the policy loss function of A3C;

[0179] The policy network model inputs the comprehensive feature vector for the hth time, and then outputs the action of mass spectrometry analysis of the hydrolyzed protein for the h+1th time;

[0180] The action of the first hydrolysis protein mass spectrometry analysis comes from manual experience or historical parameters used in similar protein mass spectrometry analysis. When h = 0, the comprehensive feature vector comes from the data after the first hydrolysis protein mass spectrometry analysis;

[0181] The policy network model continuously iterates and generates actions according to the above method. When any termination condition is reached, the step of generating actions is stopped. The termination conditions include:

[0182] The number of generated actions reaches the set upper limit; for example, 20 times;

[0183] The peptide coverage reaches the set value; for example, 96%. Generally speaking, this value needs to be set to be greater than 85%.

[0184] Step 103, execute the action output by the strategy network model, apply the action to the new protein, obtain new peptide spectrum data, and obtain the corresponding protein identification result according to the obtained new peptide spectrum data analysis.

[0185] In one embodiment of the present invention, the protein identification result is obtained based on the peptide spectrum data analysis using existing methods, and the present invention is not limited thereto.

[0186] The present invention processes data based on these optimized strategies, performs spectrum matching and protein identification, and can obtain more accurate and comprehensive identification results.

[0187] A peptide spectrum matching data analysis system for small molecule peptide donkey-hide gelatin, comprising the following modules:

[0188] A data acquisition module 100 is used to obtain historical protein identification data and a theoretical mass spectrometry protein database;

[0189] Function: Responsible for collecting and organizing necessary input data;

[0190] Submodule 1, historical data importer: extracts historical protein identification records from the database, including peptide spectrum data (peptide m / z, abundance, b / y ion information) and enzymatic hydrolysis details (enzyme type, enzyme cleavage site, enzymatic hydrolysis conditions);

[0191] Submodule 2, theoretical database interface: docking with the protein database to obtain the latest protein sequence information for comparative analysis;

[0192] The process optimization module 200 is used to construct a strategy network model and output the best enzymatic hydrolysis strategy and mass spectrometry parameter configuration that should be adopted under the current state;

[0193] Function: Build models based on historical data to optimize the identification process;

[0194] Submodule 3, Strategy Network Model Builder: Use machine learning algorithms to train models based on historical identification data to determine the best enzymatic strategy, mass spectrometry parameter configuration, and database query solution;

[0195] Submodule 4, Strategy Optimization Engine: further optimizes the strategy network model to ensure maximum identification accuracy and efficiency;

[0196] Submodule 5, feature vector constructor: integrates multi-source information to form a comprehensive feature vector as the input of the policy network;

[0197] Submodule 6, policy network model: receives feature vectors and outputs the optimal action sequence customized for the current task;

[0198] Submodule 7, Dynamic Adjustment Adapter: Converts the output of the policy network into specific operation instructions to prepare for the processing of new data;

[0199] The peptide spectrum data analysis and identification module 300 is used to obtain the corresponding protein identification result according to the obtained new peptide spectrum data analysis;

[0200] Function: Apply optimization strategies to process new data and perform protein identification;

[0201] Submodule eight, data processor: adjusts mass spectrometer parameters and processes new peptide spectrum data based on the output of the strategy network;

[0202] Submodule nine, spectrum matching and identifier: uses the optimized strategy to match the processed spectra, compares them with the theoretical database, and outputs the protein identification results;

[0203] Submodule ten, result verification and feedback system: perform quality control on the identification results and provide feedback to the previous modules to continuously optimize the model performance.

[0204] A storage medium stores non-transitory computer-readable instructions. When the non-transitory computer-readable instructions are executed by a computer, the steps in the above-mentioned method for analyzing polypeptide spectrum matching data of a small molecule peptide donkey-hide gelatin can be executed.

[0205] The above describes an embodiment of the present embodiment, but the present embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present embodiment, ordinary technicians in this field can also make many forms, all of which are within the protection of the present embodiment.

Claims

1. A method for analyzing polypeptide spectrum matching data of small molecule peptide donkey-hide gelatin, characterized in that: The following steps are involved: Step 101, obtaining historical protein identification data and a theoretical mass spectrum protein database, wherein the historical protein identification data includes peptide spectrum data and enzymatic hydrolysis information, wherein the peptide spectrum data includes the mass-to-charge ratio value and abundance of peptide segments and the mass-to-charge ratio value and abundance of b / y ions in fragmentation mode, and the enzymatic hydrolysis information includes the type of enzyme used, and each enzyme corresponds to a specific enzyme cleavage site, enzymatic hydrolysis time and conditions; Step 102, constructing a strategy network model based on historical protein identification data, the input of the strategy network model is a comprehensive feature vector, and the output is an action in the current state; The strategy network model includes: definition of state space: enzymatic hydrolysis information: type of enzyme used; peptide spectrum data: mass-to-charge ratio, abundance, mass-to-charge ratio and abundance of b / y ions in fragmentation mode; identified peptide sequence: confirmed peptide sequence and its position in protein; database query status: number of peptides matched with known databases, matching degree, and coverage of matched peptides; resource consumption: elapsed time, reagent consumption; encoding enzymatic hydrolysis information, peptide spectrum data, identified peptide sequence, database query status and resource consumption information and concatenating them into a comprehensive feature vector; definition of action space: selection of enzymatic hydrolysis strategy: selection of specific enzyme; mass spectrometry parameter adjustment: adjustment of scanning range, resolution, fragmentation mode, and fragmentation energy; database query: selection of query database and setting of search parameters; experimental process control: decision on whether to continue enzymatic hydrolysis, repeat mass spectrometry analysis, or change experimental direction; The strategy network model inputs the comprehensive feature vector for the hth time, and then outputs the action of the h+1th hydrolysis protein mass spectrometry analysis; the action of the first hydrolysis protein mass spectrometry analysis comes from manual experience or historical parameters used in similar protein mass spectrometry analysis. When h=0, the comprehensive feature vector comes from the data after the first hydrolysis protein mass spectrometry analysis; Step 103, execute the action output by the strategy network model, apply the action to the new protein, obtain new peptide spectrum data, and obtain the corresponding protein identification result according to the obtained new peptide spectrum data analysis.

2. The method for analyzing polypeptide spectrum matching data of a small molecule peptide donkey-hide gelatin according to claim 1, characterized in that: In step 101, the historical protein identification data is denoised to remove noise and low-quality peaks; the mass-to-charge ratio and abundance are normalized to ensure that the data are on a uniform scale; Peak alignment was performed on the experimental mass spectrometry data and theoretical mass spectrometry data to ensure that the peaks of the same peptide in different spectra could match; historical peptide spectrum data were extracted, and the mass-to-charge ratio, abundance and type of fragmentation products of each peptide and its fragmentation products were used as features.

3. The method for analyzing polypeptide spectrum matching data of a small molecule peptide donkey-hide gelatin according to claim 1, characterized in that: The strategy network model includes: establishing a reward function for the output action of the strategy network model. The reward goal is to select the optimal enzymatic hydrolysis strategy, mass spectrometry parameter settings, and database query methods to maximize the accuracy and efficiency of identification. The reward function is as follows: Rewards given when the identified protein exactly matches the real protein: Rewards given when the identified protein partially matches the real protein: in, Indicates the coverage of the currently identified peptides. Indicates the maximum coverage; The reward for each 1% increase in peptide coverage is: The calculation formula for resource efficiency bonus is: in, Indicates the preset resource threshold. Indicates the actual consumption of resources; The rewards deducted for incorrect identification are: For every 10% over the resource budget, the reward deducted is: The rewards deducted for repeated invalid operations are: The rewards for exploration are: The rewards for continued learning are: The total reward is calculated as follows: in, represents the total reward, represents the reward value obtained when the protein completely matches the true protein, represents the reward value obtained when the protein partially matches the real protein, Indicates the reward value obtained for every 1% increase in peptide coverage. Indicates the reward value obtained by completing the appraisal when the resources are below a certain percentage of the preset threshold. It indicates the penalty value obtained when the identified protein does not match the real one. Indicates the penalty value obtained when exceeding the preset resource threshold. Indicates the penalty value obtained when repeating invalid operations. Represents the reward value obtained by exploration, Represents the reward value obtained from continuous learning, Indicates the coverage of the currently identified peptides. represents the theoretical maximum coverage, Indicates the preset resource threshold. Indicates the actual consumption of resources. , , , , , , , , They represent the 1st, 2nd, 3rd, 4th, 5th, 6th, 7th, 8th and 9th conditional coefficients respectively.

4. The method for analyzing polypeptide spectrum matching data of a small molecule peptide donkey-hide gelatin according to claim 3, characterized in that: Strategy network model: including input layer, first hidden layer, second hidden layer and output layer; The comprehensive feature vector is input into the input layer, the input layer outputs the comprehensive feature vector to the first hidden layer to generate a first intermediate feature vector, the first hidden layer outputs the first intermediate feature vector to the second hidden layer to generate a second intermediate feature vector, the second hidden layer outputs the second intermediate feature vector to the output layer to generate an action vector, and the action vector mapping obtains the probability of performing the corresponding action in the current state.

5. The method for analyzing polypeptide spectrum matching data of a small molecule peptide donkey-hide gelatin according to claim 4, characterized in that: The calculation formula for the first hidden layer is: in, represents the first intermediate eigenvector, represents the ReLU activation function, represents the result of linear transformation of the comprehensive feature vector, represents the comprehensive feature vector, represents the first weight matrix, represents the first bias term; The calculation formula for the second hidden layer is: in, represents the second intermediate eigenvector, represents the ReLU activation function, represents the result of linear transformation of the first intermediate eigenvector, The first intermediate eigenvector, represents the second weight matrix, represents the second bias term; The calculation formula of the output layer is: in, Indicates action characteristics. represents the third weight matrix, represents the second intermediate eigenvector, represents the third bias term, represents the probability of executing the i-th action in the current state, represents the i-th component in the action feature vector, represents the jth component in the action feature vector, and , n represents the number of all executable actions, and e represents a natural constant.

6. The method for analyzing polypeptide spectrum matching data of a small molecule peptide donkey-hide gelatin according to claim 5, characterized in that: The loss function of the policy network model is the Clip Loss in PPO or the policy loss function of A3C.

7. A polypeptide spectrum matching data analysis system for small molecule peptide donkey-hide gelatin, characterized in that: The method for performing the steps in the method for analyzing the polypeptide spectrum matching data of a small molecule peptide donkey-hide gelatin as claimed in any one of claims 1 to 6 comprises the following modules: Data acquisition module, used to obtain historical protein identification data and theoretical mass spectrometry protein database; The process optimization module is used to build a strategy network model and output the best enzymatic hydrolysis strategy and mass spectrometry parameter configuration that should be adopted under the current state; The peptide spectrum data analysis and identification module is used to obtain the corresponding protein identification results based on the new peptide spectrum data analysis.

8. A storage medium storing non-transitory computer-readable instructions, characterized in that: When the non-transitory computer-readable instructions are executed by a computer, the steps in the method for analyzing polypeptide spectrum matching data of a small molecule peptide donkey-hide gelatin as claimed in claim 6 can be executed.

Images