A method for predicting the binding affinity between a drug and a protein

Abstract

The present invention belongs to the technical field of data mining, and particularly relates to a method for predicting the binding affinity between drugs and proteins, which includes: obtaining multi-modal drug-protein data and inputting it into a trained affinity prediction model to obtain a predicted value; the training process of the affinity prediction model includes: obtaining multi-modal drug-protein data and inputting it into a feature extraction module to obtain multi-modal drug-protein features; splicing the multi-modal drug-protein features to obtain an initial comprehensive feature representation; inputting the initial comprehensive feature representation into a dynamic attention module applying a memory module to obtain a final comprehensive feature; inputting the final comprehensive feature into a prediction module to obtain a prediction result; calculating a loss function value according to the prediction result, and updating the model parameters according to the loss function value; the present invention uses a historical attention weight to dynamically fuse multi-modal features through an attention weight memory mechanism, improving the performance of the prediction model.

Description

Technical Field

[0001] The present invention belongs to the technical field of machine learning and data mining, and particularly relates to a method for predicting drug-protein binding affinity. Background Art

[0002] A key step in the drug development process is to predict the binding affinity between a drug and a target protein. This process traditionally relies on experimental methods, but the experimental process is usually time-consuming and costly. Therefore, it is particularly important to develop an efficient and accurate computational model to predict drug-protein binding affinity.

[0003] In recent years, with the development of deep learning technology, more and more researchers have applied it to the prediction of drug-protein interactions. These methods have significantly improved the prediction accuracy by learning feature representations from large-scale data. Existing drug-protein binding affinity prediction methods mainly include machine learning-based and deep learning-based methods. Machine learning methods rely on manually designed features, which may not be comprehensive enough in complex molecular and protein interactions. Deep learning methods overcome the limitations of manually designed features by automatically extracting features, but most of them only use a single type of data and fail to fully utilize the multi-modal information of drugs and proteins. In addition, existing deep learning methods usually use fixed attention weights during the training process, lacking a mechanism for dynamic adjustment, resulting in suboptimal performance of the model under different data conditions.

[0004] Therefore, the existing technology has problems such as insufficient feature extraction and ineffective fusion of multi-modal data due to the lack of a dynamic adjustment mechanism. Summary of the Invention

[0005] To solve the above problems of the existing technology, the present invention adopts a method for predicting drug-protein binding affinity, which is characterized by including: obtaining multi-modal drug-protein data, inputting the multi-modal drug-protein data into a trained affinity prediction model to obtain a predicted value of drug-protein affinity; the affinity prediction model includes: a feature extraction module, a dynamic attention module applying a memory module, and a prediction module; the training process of the affinity prediction model includes:

[0006] S1: Obtain multi-modal drug-protein data, and input the multi-modal drug-protein data into the feature extraction module to obtain multi-modal drug-protein features;

[0007] S2: Concatenate the multi-modal drug-protein features to obtain an initial comprehensive feature representation;

[0008] S3: Input the initial comprehensive feature representation into the dynamic attention module of the application memory module to obtain the final comprehensive feature;

[0009] S4: Input the final comprehensive feature into the prediction module to obtain the prediction result;

[0010] S5: Calculate the loss function value according to the prediction result, update the model parameters according to the loss function value, and when the loss function value is minimized, obtain the trained affinity prediction model.

[0011] The multi-modal drug-protein data includes: the SMILES string of the drug, the molecular graph structure of the drug, and the amino acid sequence of the protein.

[0012] The feature extraction module includes: a SMILES feature extraction module, a molecular graph structure feature extraction module, and an amino acid sequence feature extraction module.

[0013] The SMILES feature extraction module includes: an embedding layer, a one-dimensional convolutional neural network, and an LSTM network; the processing of the SMILES string of the drug by the SMILES feature extraction module includes: converting the SMILES string into integer encoding; inputting the integer-encoded SMILES string into the embedding layer to obtain an embedding matrix; inputting the embedding matrix into the one-dimensional convolutional neural network to obtain local features; inputting the local features into the LSTM network to obtain the SMILES string features.

[0014] The processing of the molecular graph structure of the drug by the molecular graph structure feature extraction module includes: representing the molecular graph structure of the drug as a molecular undirected graph G=(V,E), where V is the node set, the nodes represent the atoms of the molecular graph structure, and E is the edge set, and the edges represent the chemical bonds of the molecular graph structure; initializing the feature vectors of each node and each edge; using a graph convolutional network to process the feature vectors of the nodes and edges in the molecular undirected graph G to obtain the molecular graph structure features.

[0015] The amino acid sequence feature extraction module includes: an embedding layer, a one-dimensional convolutional neural network, and an LSTM network; the processing of the amino acid sequence of the protein by the amino acid sequence feature extraction module includes; converting the amino acid sequence of the protein into integer encoding; inputting the integer-encoded amino acid sequence into the embedding layer to obtain an embedding matrix; inputting the embedding matrix into the one-dimensional convolutional neural network to obtain local features; inputting the local features into the LSTM network to obtain the amino acid sequence features.

[0016] The dynamic attention module of the application memory module includes a memory matrix, and the memory matrix contains the attention weights of multiple historical time steps; the processing of the initial comprehensive feature representation by the dynamic attention module of the application memory module includes:

[0017] Calculate the initial attention weight α of the initial comprehensive feature representation at time step t t ; According to the initial attention weight α t Update the memory matrix M at time step t-1 t-1 , to obtain the memory matrix M at time step t t , according to the initial attention weight α t and the memory matrix M t Calculate the final attention weight α at time step t t,final , according to the attention weight α t,final Adjust the initial comprehensive feature representation to obtain the final comprehensive feature.

[0018] Updating the memory matrix M t-1 includes: If the memory matrix M t-1 has already stored the attention weights of T time steps, then in the memory matrix M t-1 remove the attention weight of the earliest time step and add the attention weight α t ; Otherwise, directly add the attention weight α in the memory matrix M t-1 ; where T is the maximum dimension of the memory matrix. t

[0019] Calculating the final attention weight at time step t includes: calculating the average value of the attention weights of the historical time steps in the memory matrix M t , and performing a weighted average on the average value and the attention weight α t to obtain the final attention weight at time step t.

[0020] The prediction module includes: a KAN neural network and a multi-layer perceptron; the prediction module processes the final comprehensive feature including: inputting the final comprehensive feature representation into the KAN neural network, and inputting the output of the KAN neural network into the multi-layer perceptron to obtain a prediction result.

[0021]

[0022] Beneficial effects:

[0022] 1. In the dynamic attention module of the present invention, a memory module is added to obtain an attention weight memory mechanism. Through the attention weight memory mechanism, historical attention weights are used to dynamically fuse multi-modal features, improving the performance of the prediction model. 2. By storing the historical attention weight distribution in the memory matrix, compared with storing other historical information, the model can be made more lightweight, reducing the computational complexity and time cost of model training. 3. The attention weight memory mechanism combines the historical attention weight distribution to more flexibly adjust the current attention weight distribution. Compared with storing other historical information, it can better retain and utilize the relevance of data context to adapt to new input data. 4. The present invention makes full use of the multi-modal information of drugs and proteins, namely sequence data and graph structure data, to improve the performance of the prediction model. 5. By introducing the KAN network, the present invention can better capture the relationships between features and optimize the feature extraction ability of the model. Brief Description of the Drawings

[0023] Figure 1 It is a flowchart of a method for predicting drug-protein binding affinity provided by an embodiment of the present invention;

[0024] Figure 2 It is a structural diagram of an affinity prediction model provided by an embodiment of the present invention. Detailed Embodiments

[0025] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0026] As Figure 1 shown, the present invention proposes a method for predicting drug-protein binding affinity, including: obtaining multi-modal drug-protein data; the multi-modal drug-protein data includes: the SMILES string of the drug, the molecular graph structure of the drug, and the amino acid sequence of the protein. The multi-modal drug-protein data is input into the trained affinity prediction model to obtain the predicted value of drug-protein affinity; the affinity prediction model includes: a feature extraction module, a dynamic attention module applying a memory module, and a prediction module; the training process of the affinity prediction model includes:

[0027] S1: Collect multi-modal drug-protein data, and input the SMILES string of the drug, the molecular graph structure of the drug, and the amino acid sequence of the protein into the feature extraction module respectively to obtain SMILES string features, molecular graph structure features, and amino acid sequence features;

[0028] S2: Concatenate the SMILES string features, molecular graph structure features, and amino acid sequence features to obtain an initial comprehensive feature representation;

[0029] S3: Input the initial comprehensive feature representation into the dynamic attention module that applies the memory module to obtain the final comprehensive feature;

[0030] S4: Input the final comprehensive feature into the prediction module to obtain a prediction result;

[0031] S5: Calculate the loss function value based on the prediction result, update the model parameters according to the loss function value, and when the loss function value is minimized, obtain the trained model.

[0032] As Figure 2 shown, the feature extraction module includes: a SMILES feature extraction module, a molecular graph structure feature extraction module, and an amino acid sequence feature extraction module;

[0033] Furthermore, the SMILES feature extraction module includes: an embedding layer, a one-dimensional convolutional neural network, and an LSTM network; The processing of the drug SMILES string includes:

[0034] Step 1: Convert the SMILES string to integer encoding (Label encoding);

[0035] Step 2: Input the integer-encoded SMILES string into the embedding layer (Embedding layer) to convert the discrete characters into a continuous vector representation, that is, an embedding matrix.

[0036] Step 3: Input the embedding matrix into a one-dimensional convolutional neural network (1D-CNN) to extract local features;

[0037] Step 4: Input the output of the 1D-CNN into the LSTM to capture the long-term dependencies in the sequence and obtain the SMILES string features:

[0038] h t , c t = LSTM(C t , h t-1 , c t-1 )

[0039] where C t is the output of the 1D-CNN, h t and c t are the hidden state and cell state of the LSTM at time step t respectively, and h t-1 and c t-1They are the hidden state and cell state of the LSTM at time step t-1 respectively. The final output of the LSTM is the SMILES string feature F SMILES 。

[0040] Furthermore, the amino acid sequence feature extraction module includes: an embedding layer, a one-dimensional convolutional neural network, and an LSTM network; the processing of the amino acid sequence of the protein by the amino acid sequence feature extraction module includes;

[0041] Step 1: Perform integer or label encoding on the amino acid sequence of the protein; for protein sequences, 20 different classes are used to represent amino acids;

[0042] Step 2: Input the encoded amino acid sequence into the embedding layer to obtain an embedding matrix;

[0043]

[0044] where L P is the length of the amino acid sequence of the protein, and F e is the embedding dimension.

[0045] Step 3: Input the embedding matrix into a one-dimensional convolutional neural network 1D-CNN to extract local features;

[0046] Step 4: Input the features extracted by 1D-CNN into the LSTM to capture the long-term dependencies of the sequence. The final output of the LSTM is the amino acid sequence feature of the protein.

[0047] Furthermore, the processing of the molecular graph structure of the drug by the molecular graph structure feature extraction module includes:

[0048] Step 1: Molecular Graph Representation: Represent the molecular graph structure of the drug as a molecular undirected graph G=(V, E), where V is the set of nodes, the nodes represent the atoms of the molecular graph structure, and E is the set of edges, and the edges represent the chemical bonds of the molecular graph structure;

[0049] Step 2 Feature Attribution: Initialize the feature vector of each node according to the type and attributes of the atom, and initialize the feature vector of each edge according to the type and attributes of the chemical bond;

[0050] Step 3: Use a graph convolutional network (GCN) to process the feature vectors of the nodes and edges in the molecular undirected graph G to obtain the molecular graph structure feature; the specific process is as follows:

[0051] Perform graph convolution operations on each node to update the node feature vector; the graph convolution operation can be expressed as:

[0052]

[0053] Among them, N(v) is the set of neighbor nodes of node v, w is a neighbor node of node v, and c vw is the normalization coefficient, and W (k) is the weight matrix of the k-th layer of graph convolution, is the feature representation of node w output by the k-th layer of graph convolution, is the feature representation of node v output by the (k + 1)-th layer of graph convolution, and b (k) is the bias term of the k-th layer of graph convolution, and σ is the activation function;

[0054] Finally, the final feature representation of each node output by the last layer of graph convolution is obtained

[0055] Through global pooling (Pooling), the final feature representation is aggregated into a vector matrix with a dimension of d2×n to obtain the molecular graph structure feature F Graph :

[0056]

[0057] Furthermore, the multi-modal feature fusion and splicing process includes:

[0058] Perform a splicing operation on the features of each modality to form an initial comprehensive feature representation:

[0059] F concat = [F SMILES ||F Protein ||F Graph

[0060] Among them, represents the SMILES string feature, represents the protein amino acid sequence feature, represents the molecular graph structure feature, || represents the vector splicing operation, d1, d2, d3, and n are the dimensions of the above three features, and the initial comprehensive feature representation after splicing is F concat ∈R m×n , and m = d1 + d2 + d3.

[0061] Furthermore, applying the dynamic attention module (Attention block) of the memory module (Memory Module) to process the initial comprehensive feature representation includes:

[0062] Step 1: Calculate the attention weight α of the initial comprehensive feature representation at the current time step t t , specifically including:

[0063] ​Calculate the attention scores of the initial comprehensive feature representation at the current time step t:

[0064]

[0065] where, e t represents the attention score of the feature vector at the current time step t, represents the weight matrix used to calculate the attention score, F concat,t represents the feature vector in the comprehensive feature representation at the current time step t, that is, the t-th feature vector of the comprehensive feature representation, b a is the bias term.

[0066] Use the Softmax function to convert the attention scores at the current time step t into the attention weights at the current time step t:

[0067]

[0068] where, α t represents the attention weight of the feature vector at the current time step t, and m is the number of feature vectors of the initial comprehensive feature representation.

[0069] Step 2: Update the memory matrix M at the previous time step t - 1 according to the attention weight α t at the current time step t, specifically including: t-1

[0070] The memory matrix M records the attention weight distribution of the most recent T historical time steps, which is expressed as follows:

[0071] M ∈ R T×n

[0072] where, n represents the dimension of the attention weight vector at each time step.

[0073] If the memory matrix M at the previous time step t-1 has already stored the attention weights of the most recent T time steps, then remove the attention weights of the earliest time step in the memory matrix M t-1 and add the attention weight α t at the current time step. The formula is as follows:

[0074] M t = [α t || M t-1,1:T-1

[0075] Otherwise, directly add the attention weight α t-1 at the current time step to the memory matrix M t ; where, M t represents the memory matrix at the current time step t, and α tDenote the attention weights at the current time step, M t-1,1:T-1 Denote the result of removing the earliest record from the memory matrix at the previous time step.

[0076] Step 3: According to the attention weights α at the current time step t and the memory matrix M t Calculate the final attention weights at time step t, and adjust the initial comprehensive feature representation according to the final attention weights at time step t to obtain the final comprehensive features;

[0077] Dynamically adjust the representation of each feature vector according to the current attention weights and the historical information in the memory matrix. The specific steps are as follows:

[0078] Perform a weighted average of the attention weights at the current time step and the historical weights in the memory matrix to form the final attention weights. The specific formula is as follows:

[0079]

[0080] where α t,final denotes the final attention weights at time step t, γ denotes a parameter that controls the ratio of the current attention weights to the historical weights, and its value range is [0, 1]. α t denotes the attention weights at the current time step, and M t,k denotes the attention weights at the k-th historical time step in the memory matrix.

[0081] Adjust the initial comprehensive feature representation using the final attention weights:

[0082]

[0083] where F Attention denotes the final comprehensive feature representation, α t,final denotes the final attention weights of the feature vector at time step t, and F Concat,t denotes the t-th feature vector in the comprehensive feature representation.

[0084] Furthermore, the prediction module includes: a KAN neural network and a multi-layer perceptron; the processing of the final comprehensive features by the prediction module includes:

[0085] Step 1: Input the final comprehensive feature representation into the KAN neural network;

[0086] KAN (Knowledge-Aware Neural Network) optimizes the feature extraction ability of the model by introducing domain knowledge:

[0087] H KAN = KAN(F Attention )

[0088] Among them, H KAN is the feature representation extracted by the KAN neural network. The structure of the KAN neural network can include multiple layers, such as graph convolutional layers, graph attention layers, etc., which are used to better capture the relationships between features.

[0089] Step 2: Input the output of the KAN neural network into a multi-layer perceptron;

[0090] The multi-layer perceptron MLP is used to further process the feature representation H extracted by KAN KAN , and finally obtain the prediction result.

[0091] The MLP contains several fully connected layers and activation functions. The output of each layer can be expressed as:

[0092]

[0093] Among them, represents the output of the l-th layer of the MLP, W (l) represents the weight matrix of the l-th layer, b (l) represents the bias term of the l-th layer, and σ represents the activation function.

[0094] The output layer of the final MLP is

[0095]

[0096] Among them, O MLP represents the final output (Output) of the MLP, Y pred is the predicted drug-protein binding affinity value, W out represents the weight matrix of the output layer, b out represents the bias term of the output layer, and L represents the number of layers of the MLP.

[0097] Furthermore, the loss function L of the affinity prediction model MSE is;

[0098]

[0099] Among them, y i is the true drug-protein affinity value of the i-th sample, is the affinity value predicted by the model, and N is the number of samples.

[0100] The above-mentioned embodiments further illustrate the purpose, technical solutions and advantages of the present invention. It should be understood that the above-mentioned embodiments are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made to the present invention within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for predicting the binding affinity between a drug and a protein, characterized in that, Including: Obtain multi-modal drug-protein data, input the multi-modal drug-protein data into a trained affinity prediction model, and obtain a predicted value of the drug-protein affinity; The affinity prediction model includes: a feature extraction module, a dynamic attention module applying a memory module, and a prediction module; the training process of the affinity prediction model includes: S1: Obtain multi-modal drug-protein data, input the multi-modal drug-protein data into the feature extraction module, and obtain multi-modal drug-protein features; S2: Concatenate the multi-modal drug-protein features to obtain an initial comprehensive feature representation; S3: Input the initial comprehensive feature representation into the dynamic attention module applying the memory module to obtain a final comprehensive feature; S4: Input the final comprehensive feature into the prediction module to obtain a prediction result; S5: Calculate the loss function value according to the prediction result, update the model parameters according to the loss function value, and when the loss function value is the smallest, obtain a trained affinity prediction model; The multi-modal drug-protein data includes: the SMILES string of the drug, the molecular graph structure of the drug, and the amino acid sequence of the protein; The feature extraction module includes: a SMILES feature extraction module, a molecular graph structure feature extraction module, and an amino acid sequence feature extraction module; The SMILES feature extraction module includes: an embedding layer, a one-dimensional convolutional neural network, and an LSTM network; the processing of the SMILES string of the drug by the SMILES feature extraction module includes: converting the SMILES string into an integer encoding; inputting the integer-encoded SMILES string into the embedding layer to obtain an embedding matrix; inputting the embedding matrix into the one-dimensional convolutional neural network to obtain local features; inputting the local features into the LSTM network to obtain SMILES string features; The processing of the molecular graph structure of the drug by the molecular graph structure feature extraction module includes: representing the molecular graph structure of the drug as a molecular undirected graph G=(V, E), where V is the set of nodes, the nodes represent the atoms of the molecular graph structure, and E is the set of edges, the edges represent the chemical bonds of the molecular graph structure; initializing the feature vectors of each node and each edge; using a graph convolutional network to process the feature vectors of the nodes and edges in the molecular undirected graph G to obtain molecular graph structure features; The amino acid sequence feature extraction module includes: an embedding layer, a one-dimensional convolutional neural network, and an LSTM network; the processing of the amino acid sequence of the protein by the amino acid sequence feature extraction module includes; converting the amino acid sequence of the protein into an integer encoding; inputting the integer-encoded amino acid sequence into the embedding layer to obtain an embedding matrix; inputting the embedding matrix into the one-dimensional convolutional neural network to obtain local features; inputting the local features into the LSTM network to obtain amino acid sequence features.

2. The method for predicting the drug-protein binding affinity according to claim 1, wherein The dynamic attention module applying the memory module includes a memory matrix, and the memory matrix contains the attention weights of multiple historical time steps; The processing of the initial comprehensive feature representation by the dynamic attention module applying the memory module includes: Calculate the initial attention weight α of the initial comprehensive feature representation at time step t t ; According to the initial attention weight α t Update the memory matrix M at time step t - 1 t-1 to obtain the memory matrix M at time step t t According to the initial attention weight α t and the memory matrix M t calculate the final attention weight α at time step t t,final According to the attention weight α t,final adjust the initial comprehensive feature representation to obtain the final comprehensive feature.

3. The method for predicting the drug-protein binding affinity according to claim 2, wherein Update the memory matrix M t-1 including: if the memory matrix M t-1 has stored the attention weights for T time steps, then remove the attention weights of the earliest time step in the memory matrix M t-1 and add the attention weight α t ; otherwise, directly add the attention weight α t-1 to the memory matrix M t ; where T is the maximum dimension of the memory matrix.

4. A method for predicting the drug-protein binding affinity according to claim 2, wherein Calculating the final attention weights at time step t includes: calculating the memory matrix M t the average of the attention weights of historical time steps in t , and performing a weighted average on the average and the attention weight α t to obtain the final attention weights at time step t.

5. A method for predicting the binding affinity between a drug and a protein according to claim 1, characterized in that, The prediction module includes: a KAN neural network and a multi-layer perceptron; the processing of the final comprehensive feature by the prediction module includes: inputting the representation of the final comprehensive feature into the KAN neural network, and inputting the output of the KAN neural network into the multi-layer perceptron to obtain a prediction result.

Images