A polypeptide drug toxicity prediction method and system based on artificial intelligence
By employing an AI-based peptide toxicity prediction method that utilizes Bayesian regression and convolutional neural network models, the problem of time-consuming and costly traditional peptide drug toxicity detection has been solved, enabling accurate prediction and rapid screening of peptide toxicity.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- THE THIRD AFFILIATED HOSPITAL OF SOUTHERN MEDICAL UNIV (ACAD OF ORTHOPEDICS GUANGDONG PROVINCE)
- Filing Date
- 2025-03-10
- Publication Date
- 2026-05-12
AI Technical Summary
Traditional methods for detecting the toxicity of peptide drugs are time-consuming and costly, making it difficult to predict the toxicity of peptide drugs quickly and effectively, especially considering the uncertainty of hydrolysis products generated during their metabolism in vivo.
An artificial intelligence-based approach was adopted, using a Bayesian regression model to construct a peptide metabolic rate constant prediction model, and a convolutional neural network to construct a peptide toxicity prediction model. The toxicity scores of peptides and their hydrolysis products were predicted by utilizing peptide sequence features and amino acid semantic relationships.
It enables accurate prediction of peptide toxicity, improves prediction efficiency and accuracy, reduces costs, and simplifies the drug screening process.
Smart Images
Figure CN120183537B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of drug screening, and in particular to an artificial intelligence-based method and system for predicting the toxicity of peptide drugs. Background Technology
[0002] Drug screening is a complex and time-consuming process, with toxicity assessment being a crucial step. Traditional compound toxicity testing techniques typically require biochemical assays, cell experiments, and even animal models, methods that are not only time-consuming but also costly. A drug's toxicity determines whether it can pass experimental and regulatory review; therefore, rapid and effective identification of peptide toxicity is a critical step in peptide drug development. However, due to the complexity and diversity of peptides, and the uncertainty of their hydrolysis products generated during in vivo metabolism, this predictive task remains challenging. Summary of the Invention
[0003] To address the technical problems existing in the prediction of peptide drug toxicity in the prior art, this invention provides a method and system for predicting peptide drug toxicity based on artificial intelligence.
[0004] This invention is achieved through the following technical solution:
[0005] An artificial intelligence-based method for predicting peptide drug toxicity includes:
[0006] S1: Collect sequence data and properties of polypeptide molecules; the sequence data and properties of polypeptide molecules include the sequence of polypeptide molecules, related biological activity data, physicochemical property parameters, known toxicity information, and possible metabolite data;
[0007] S2: Construct a rate constant prediction model for peptide metabolism based on a Bayesian regression model;
[0008] S3: Construct a toxicity prediction model to predict the toxicity score of peptide molecules and their hydrolysis products, including extracting fragment sequences from peptide sequences to generate sequence features of peptides and peptide fragments; inputting the selected feature vectors into the trained toxicity prediction model to predict peptide toxicity, constructing a kinetic simulation peptide toxicity prediction model, and obtaining the toxicity prediction results of peptide molecules and their hydrolysis products.
[0009] Furthermore, in step S2, the rate constant prediction model for peptide metabolism is constructed to establish the relationship between molecular weight, hydrophobicity, molecular fragment descriptor, charge distribution, number of hydrogen bond donors and acceptors, and the metabolic rate constant of the peptide and its metabolites.
[0010] Furthermore, step S2 also includes feature selection and importance assessment. By using feature importance assessment, key features in the prediction of metabolic rate constant are screened out, and the contribution of each feature to the model prediction is quantified.
[0011] Furthermore, step S3 also includes:
[0012] S31: Data preprocessing, including acquiring drug datasets and extracting molecular weight, hydrophobicity, molecular fragment descriptors, charge distribution, peptide sequences, peptide toxicity data, metabolite data, and the number of peptide secondary structures by processing raw biochemical test data.
[0013] S32. Perform word segmentation on the polypeptide sequence, use a sliding window of length T to obtain polypeptide fragments, count the frequency of fragments to form a list of key fragments, and construct the feature space of the sequence.
[0014] S33. The amino acid sequence is characterized by word embedding model, each amino acid is mapped to a high-dimensional vector, the semantic relationship and interaction between amino acids are captured, and the feature vector of the polypeptide sequence is obtained.
[0015] S34. Based on kinetic modeling, calculate the rate of change of concentration over time, and use a convolutional neural network to construct a peptide toxicity prediction model to establish the relationship between peptides and toxicity.
[0016] Furthermore, when calculating the rate of change of concentration over time based on kinetic modeling, it is assumed that the metabolic pathway of peptide P in vivo follows first-order reaction kinetics, and the rate constant is predicted by the model in S2, which is expressed as follows:
[0017] i=1,2,…,n
[0018] Among them, [p] and The concentrations of the polypeptide and its metabolites, respectively, and k and kJ are given. These are the rate constants of the polypeptide and its metabolite, respectively. The rate at which the peptide concentration [p] changes with time t. For the concentration of metabolites The rate at which it changes over time.
[0019] At time t, the concentration of the metabolite is = Peak time is The peak concentration is obtained by back-substituting the formula. .
[0020] Furthermore, the method of constructing a peptide toxicity prediction model using a convolutional neural network, and establishing the relationship between peptides and toxicity, includes: based on the relationship between metabolite concentration and toxicity, the toxicity contribution can be expressed as:
[0021]
[0022] in, This represents the peak concentration of the metabolite. For model weights, is a non-linear activation function, and i is the index of the polypeptide metabolite.
[0023] Furthermore, step S34 also includes dividing the dataset in S31 into a training set, a validation set, and a test set for model training, parameter tuning, and evaluation. The labeled peptide toxicity training set is used for training to learn the sequence characteristics and combinatorial relationships of peptide toxicity.
[0024] Furthermore, step S34 also includes evaluating the trained model using a validation set after training is completed, evaluating the model's performance on unseen data using a test set, verifying its accuracy and generalization ability in predicting peptide toxicity, and optimizing and improving the model based on the results analysis.
[0025] This invention also provides an artificial intelligence-based peptide drug toxicity prediction system, based on the aforementioned artificial intelligence-based peptide drug toxicity prediction method, comprising:
[0026] The data acquisition module is used to acquire sequence data and property tags of peptide molecules and identify sequence combination features in peptide drug molecules.
[0027] A rate constant prediction module for peptide metabolism, which is used to predict the rate constant of peptide metabolism;
[0028] The toxicity prediction module is used to extract fragment sequences from peptide sequences, generate sequence features of peptides and peptide fragments, input the selected feature vectors into a trained toxicity prediction model to predict drug properties, construct a machine learning model for kinetic simulation, and obtain toxicity prediction results of peptide molecules and their hydrolysis products.
[0029] Furthermore, to achieve the above objectives, the present invention also provides a computer-readable storage medium storing program instructions for an artificial intelligence-based peptide drug toxicity prediction method. The program instructions for the artificial intelligence-based peptide drug toxicity prediction method can be executed by one or more processors to implement the steps of the artificial intelligence-based peptide drug toxicity prediction method as described above.
[0030] Compared with the prior art, the beneficial effects of the present invention are as follows:
[0031] By combining intelligent word segmentation and contextual statistical information, along with word embedding vectors and dynamic modeling of peptides, the system fully extracts peptide information and leverages the advantages of both to more accurately predict peptide toxicity. Furthermore, by predicting overall toxicity, corresponding amino acid combinations are extracted from contextual word frequency information, making the prediction results for toxic fragments even more precise. Attached Figure Description
[0032] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:
[0033] Figure 1 This is a schematic flowchart of an artificial intelligence-based peptide drug toxicity prediction method according to an embodiment of this application;
[0034] Figure 2 This is a schematic diagram of dynamic input and segment prediction according to an embodiment of this application. Detailed Implementation
[0035] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.
[0036] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.
[0037] It should also be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. The illustrations only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the shape, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.
[0038] See Figure 1 An artificial intelligence-based method for predicting the toxicity of peptide drugs includes the following steps:
[0039] S1: Collect sequence data and property tags of peptide molecules to ensure the accuracy and completeness of the data;
[0040] The sequence data and properties of the polypeptide molecule include the polypeptide molecule sequence, related biological activity data, physicochemical property parameters, known toxicity information, and possible metabolite data.
[0041] S2: Construct a rate constant prediction model for peptide metabolism, including:
[0042] S21. Feature Extraction: Extract the physicochemical properties of molecules using the molecular feature generation tool RDKit, including molecular weight, hydrophobicity, molecular fragment descriptors, charge distribution, number of hydrogen bond donors and acceptors;
[0043] S22. Model Construction: Using a Bayesian regression model, the relationship between molecular weight, hydrophobicity, molecular fragment descriptors, charge distribution, number of hydrogen bond donors and acceptors, and the metabolic rate constants of the peptide and its metabolites is constructed. For the regression task of the rate constant, the optimization objective is to minimize the mean squared error, which is mathematically expressed as:
[0044]
[0045] Where N is the sample size. Let be the true rate constant of the i-th sample. Let be the predicted value for the i-th sample.
[0046] S23. Feature Selection and Importance Assessment: Key features in the metabolic rate constant prediction are selected using feature importance assessment. SHAP analysis is used to quantify the contribution of each feature to the model's prediction, identify key features, and improve the model's interpretability.
[0047] S24. Calculate the rate constant through Bayesian regression to provide input parameters for dynamic modeling.
[0048] S3: Construct a toxicity prediction model to predict the toxicity fraction of peptide molecules and their hydrolysis products, including:
[0049] Fragment sequences are extracted from peptide sequences to generate sequence features of the peptide and peptide fragments. The selected feature vectors are then input into a trained toxicity prediction model to predict peptide toxicity. A kinetic simulation-based peptide toxicity prediction model is constructed to obtain the toxicity prediction results of the peptide molecule and its hydrolysis products. Specifically, the steps include:
[0050] S31: Data Preprocessing: Obtain the drug dataset and extract molecular weight, hydrophobicity, molecular fragment descriptors, charge distribution, peptide sequence, peptide toxicity data, metabolite data, and number of peptide secondary structures by processing the raw biochemical test data.
[0051] S32. Perform word segmentation on the polypeptide sequence, using a sliding window of length T to obtain polypeptide fragments, count the frequency of fragments to form a list of key fragments, and construct the sequence feature space. This process captures information in the polypeptide sequence by extracting global and local sequence features. Optionally, T can be set to 3.
[0052] It also includes segmenting the sequence into words and statistically analyzing high-frequency sequences. The core process is as follows:
[0053] a. Traverse each position in the sequence and sample using a window of length T, using a sliding window sampling from the N-terminus to the C-terminus of the polypeptide sequence;
[0054] b. Generate candidate field information, store the candidate fields, and perform statistical analysis;
[0055] c. Repeat the above operations for new sequences, maintaining a dictionary of statistical information, until there are no new sequences in the training set.
[0056] d. Set a threshold, remove fragment combinations that appear less than the threshold from the dictionary, and collect probability information.
[0057] S33. The amino acid sequence is characterized by word embedding model, which maps each amino acid to a high-dimensional vector to capture the semantic relationship and interaction between amino acids and obtain the feature vector of the polypeptide sequence.
[0058] It also includes: combining the S32 word segmentation method, generating m-2 sub-sequences with a peptide sequence length of n and a window length of 3, generating a latent representation of length 3, and padding the latent representation of the sub-sequences with 0s to make the sequence length n. The latent representations of the segmented sub-sequences and the latent representations of the peptide sequence are concatenated, so that the model can simultaneously obtain the features of the peptide sequence and peptide fragments when extracting peptide sequence information.
[0059] S34. Calculate the rate of change of concentration over time based on kinetic modeling. Assume that the metabolic pathway of peptide P in vivo follows first-order reaction kinetics, and the rate constant is predicted by the model in S2, which is expressed as follows:
[0060] i=1,2,…,n
[0061] Among them, [p] and The concentrations of the polypeptide and its metabolites, respectively, and k and kJ are given. These are the rate constants of the polypeptide and its metabolite, respectively. The rate at which the peptide concentration [p] changes with time t. For the concentration of metabolites The rate at which it changes over time.
[0062] At time t, the concentration of the metabolite is = Peak time is The peak concentration is obtained by back-substituting the formula. .
[0063] Furthermore, the method of constructing a peptide toxicity prediction model using a convolutional neural network, and establishing the relationship between peptides and toxicity, includes: based on the relationship between metabolite concentration and toxicity, the toxicity contribution can be expressed as:
[0064]
[0065] in, This represents the peak concentration of the metabolite. For model weights, is a non-linear activation function, and i is the index of the polypeptide metabolite.
[0066] The dataset in S31 is divided into training, validation, and test sets for model training, parameter tuning, and evaluation. The labeled peptide toxicity training set is used for training to learn the sequence characteristics and combinatorial relationships of peptide toxicity.
[0067] After training, the trained model is evaluated using the validation set and its performance on unseen data is evaluated using the test set. This verifies the model's accuracy and generalization ability in predicting peptide toxicity. Based on the results, the model is then tuned and improved.
[0068] In step S34, the toxicity fractions of peptides and peptide hydrolysis products were predicted using a prediction model. The word frequency information obtained in step 2 was combined with the prediction results of the toxicity fractions to deduce the combined fragments that cause toxicity.
[0069] In this embodiment, the present invention combines intelligent word segmentation and contextual statistical information with the word embedding vector of the peptide. This allows for the full extraction of peptide information and leverages the advantages of both methods, resulting in more accurate prediction of peptide toxicity. Furthermore, by predicting overall toxicity, corresponding amino acid combinations are extracted from contextual word frequency information, making the prediction results for toxic fragments even more precise.
[0070] This invention also proposes an artificial intelligence-based peptide drug toxicity prediction system, based on the artificial intelligence-based peptide drug toxicity prediction method described above, comprising:
[0071] The data acquisition module is used to acquire sequence data and property tags of peptide molecules and identify sequence combination features in peptide drug molecules.
[0072] A rate constant prediction module for peptide metabolism, which is used to predict the rate constant of peptide metabolism;
[0073] The toxicity prediction module is used to extract fragment sequences from peptide sequences, generate sequence features of peptides and peptide fragments, input the selected feature vectors into a trained toxicity prediction model to predict drug properties, construct a machine learning model for kinetic simulation, and obtain toxicity prediction results of peptide molecules and their hydrolysis products.
[0074] Furthermore, this embodiment of the invention also proposes a computer-readable storage medium storing program instructions for an artificial intelligence-based peptide drug toxicity prediction method. These program instructions can be executed by one or more processors to implement the steps of the artificial intelligence-based peptide drug toxicity prediction method as described above.
[0075] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.
Claims
1. A method for predicting the toxicity of peptide drugs based on artificial intelligence, characterized in that, include: S1: Collect sequence data and properties of polypeptide molecules; the sequence data and properties of polypeptide molecules include the sequence of polypeptide molecules, related biological activity data, physicochemical property parameters, known toxicity information, and possible metabolite data; S2: Construct a rate constant prediction model for peptide metabolism based on a Bayesian regression model; S3: Construct a kinetic simulation-based peptide toxicity prediction model to predict the toxicity fraction of peptide molecules and their hydrolysis products. This includes extracting fragment sequences from peptide sequences to generate sequence features of peptides and peptide fragments. The selected feature vectors are input into the trained peptide toxicity prediction model. Based on the kinetic simulation, a peptide prediction model is constructed to obtain the toxicity prediction results of peptide molecules and their hydrolysis products. In step S3, the construction of the peptide toxicity prediction model based on the kinetic simulation includes: S34, calculating the rate of change of concentration over time based on kinetic modeling, constructing a peptide toxicity prediction model using a convolutional neural network, and establishing the relationship between peptides and toxicity. When calculating the rate of change of concentration over time based on kinetic modeling, it is assumed that the metabolic pathway of peptide P in vivo follows first-order reaction kinetics, and the rate constant is predicted by the S2 model, which is expressed as follows: i=1,2,…,n; Among them, [p] and The concentrations of the polypeptide and its metabolites, respectively, and k and kJ are given. These are the rate constants of the polypeptide and its metabolite, respectively. The rate at which the peptide concentration [p] changes with time t. For the concentration of metabolites The rate of change over time, at time t, is the concentration of the metabolite. = Peak time is Peak concentration .
2. The method for predicting peptide drug toxicity based on artificial intelligence according to claim 1, characterized in that, In step S2, the rate constant prediction model for peptide metabolism is constructed to establish the relationship between molecular weight, hydrophobicity, molecular fragment descriptor, charge distribution, number of hydrogen bond donors and acceptors, and the metabolic rate constant of the peptide and its metabolites.
3. The method for predicting peptide drug toxicity based on artificial intelligence according to claim 2, characterized in that, Step S2 further includes feature selection and importance assessment. By using feature importance assessment, key features in the prediction of metabolic rate constant are screened out, and the contribution of each feature to the model prediction is quantified.
4. The method for predicting peptide drug toxicity based on artificial intelligence according to claim 1, characterized in that, Step S3 further includes: S31: Data preprocessing, including acquiring drug datasets and extracting molecular weight, hydrophobicity, molecular fragment descriptors, charge distribution, peptide sequences, peptide toxicity data, metabolite data, and the number of peptide secondary structures by processing raw biochemical test data. S32. Perform word segmentation on the polypeptide sequence, use a sliding window of length T to obtain polypeptide fragments, count the frequency of fragments to form a list of key fragments, and construct the feature space of the sequence. S33. The amino acid sequence is characterized by word embedding model, which maps each amino acid to a high-dimensional vector to capture the semantic relationship and interaction between amino acids and obtain the feature vector of the polypeptide sequence.
5. The method for predicting peptide drug toxicity based on artificial intelligence according to claim 4, characterized in that, The method of constructing a peptide toxicity prediction model using a convolutional neural network, and establishing the relationship between peptides and toxicity, includes: based on the relationship between metabolite concentration and toxicity, the toxicity contribution can be expressed as: ； in, This represents the peak concentration of the metabolite. For model weights, is a non-linear activation function, and i is the index of the polypeptide metabolite.
6. The method for predicting peptide drug toxicity based on artificial intelligence according to claim 4, characterized in that, Step S34 further includes dividing the dataset in S31 into a training set, a validation set, and a test set for model training, parameter tuning, and evaluation. The labeled peptide toxicity training set is used for training to learn the sequence characteristics and combinatorial relationships of peptide toxicity.
7. The method for predicting peptide drug toxicity based on artificial intelligence according to claim 4, characterized in that, Step S34 further includes evaluating the trained model using a validation set after training is completed, evaluating the model's performance on unseen data using a test set, verifying its accuracy and generalization ability in predicting peptide toxicity, and optimizing and improving the model based on the results analysis.
8. An artificial intelligence-based peptide drug toxicity prediction system, based on the artificial intelligence-based peptide drug toxicity prediction method as described in any one of claims 1 to 7, comprising: The data acquisition module is used to acquire sequence data and property tags of peptide molecules and identify sequence combination features in peptide drug molecules. A rate constant prediction module for peptide metabolism, which is used to predict the rate constant of peptide metabolism; The toxicity prediction module is used to extract fragment sequences from peptide sequences, generate sequence features of peptides and peptide fragments, input the selected feature vectors into a trained toxicity prediction model to predict drug properties, construct a machine learning model for kinetic simulation, and obtain toxicity prediction results of peptide molecules and their hydrolysis products.
9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores program instructions for an artificial intelligence-based peptide drug toxicity prediction method, which can be executed by one or more processors to implement the steps of the artificial intelligence-based peptide drug toxicity prediction method as described in any one of claims 1 to 7.