Parameter fine adjustment method for interpretability of drug molecular map structure

Abstract

The invention belongs to the field of graph neural network interpretation, and relates to an interpretable parameter fine tuning method for a drug molecular graph structure, which comprises the following steps: acquiring drug molecular graph structure data, decomposing the data into a plurality of sub-graphs, and embedding the plurality of sub-graphs into a space; adopting a greedy strategy to process the sub-graphs of the embedded space to generate a candidate semantic sub-graph set; inputting the candidate semantic sub-graph set into a graph classification model to obtain a final semantic sub-graph; generating a k-order induction graph for each node in the drug molecular graph structure data; setting a similarity threshold value, calculating the similarity between the k-order induction graph and the final semantic subgraph, and taking a structure formed by all nodes with the similarity greater than the similarity threshold value as a key structure; according to the key structure, molecular weight parameters in the interpretation process of the drug molecular map structure are adjusted through the trained multi-layer perceptron, and an optimal interpretability result is obtained; according to the method, an interpretation result which is easier to understand can be found, and personnel in related fields are assisted to make decisions.

Description

Technical Field

[0001] The present invention belongs to the field of graph neural network interpretation, and specifically relates to a parameter fine-tuning method for the interpretability of drug molecule graph structures. Background Art

[0002] Graph neural networks have been widely used in social network analysis, recommendation systems, bioinformatics and other fields due to their excellent performance in processing graph-structured data. These applications take advantage of the characteristics of graphs and can effectively capture the complex relationships between nodes. However, the "black box" characteristics of graph neural network models lead to a lack of transparency in their decision-making process, which limits their practical application in high-trust scenarios. The core of interpretability lies in identifying subgraphs in the input graph that affect the decision. This identification includes not only the nodes and edges that directly affect the decision, but also their context and association relationships. To achieve this goal, researchers have proposed a variety of methods, such as an interpretability framework based on an attention mechanism, a subgraph selection algorithm, and an explanation strategy based on counterfactuals. These methods can help users understand the behavior of the model more deeply, thereby providing effective explanations for specific tasks.

[0003] Accurate interpretation of graph neural network models not only helps people understand the model's predictions, but also reveals potential defects and directions for improvement. For example, by analyzing the interpretation results of the model, researchers can find that certain specific features may lead to incorrect predictions, which provides a basis for subsequent model optimization. At the same time, good interpretability can also promote users' trust in the model and promote the widespread application of graph neural network models in key areas.

[0004] The drug molecule graph structure, chemical composition and spatial arrangement of the drug are crucial to the biological activity and pharmacokinetic properties of the drug. How to use a graph neural network model to provide an understandable description of each chemical bond in the drug molecule graph structure, thereby improving users' effective understanding of the properties of drug molecules, is a technical problem that needs to be solved urgently. Summary of the invention

[0005] In order to solve the problems existing in the above-mentioned prior art, the present application proposes a parameter fine-tuning method for the interpretability of drug molecular graph structure, which method includes: obtaining drug molecular graph structure data; decomposing the drug molecular graph structure data into multiple subgraphs, and embedding the multiple subgraphs into space; processing the subgraphs embedded in the space using a greedy strategy, and iteratively generating a set of candidate semantic subgraphs through a score function; inputting the set of candidate semantic subgraphs into a trained graph classification model, and selecting the top R subgraphs that maximize the probability of target class prediction value as the final semantic subgraph; generating a k-order induced graph for each node in the drug molecular graph structure data; setting a similarity threshold, calculating the similarity between the k-order induced graph and the final semantic subgraph, and taking the structure composed of all nodes whose similarities are greater than the similarity threshold as the key structure; training a multilayer perceptron using the key structure; and adjusting the molecular weight parameters in the process of interpreting the drug molecular graph structure using the trained multilayer perceptron to obtain the optimal interpretability result.

[0006] Beneficial effects of the present invention:

[0007] The present invention uses the induced graph of the semantic subgraph and the original graph node to calculate the similarity, locates the key structure in the molecular graph, trains the multi-layer perceptron with the multi-hot encoding of the located key structure as the target, and uses the multi-layer perceptron to adjust the interpretation behavior of the graph neural network interpretation model, thereby improving the perception effect; the present invention can be applied to a variety of graph neural network interpretation models to enhance the comprehensibility of their interpretations; the present invention can find more understandable interpretation results to assist people in related fields in making decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 It is the overall flow chart of the present invention;

[0009] Figure 2 This is a diagram of the framework structure of the graph classification model of the present invention;

[0010] Figure 3 This is a diagram of the multi-layer perceptron framework of the present invention. DETAILED DESCRIPTION

[0011] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0012] A parameter fine-tuning method for interpretability of drug molecular graph structure, the method comprising: obtaining drug molecular graph structure data; decomposing the drug molecular graph structure data into multiple subgraphs, and embedding the multiple subgraphs into space; processing the subgraphs in the embedded space using a greedy strategy, and iteratively generating a candidate semantic subgraph set through a score function; inputting the candidate semantic subgraph set into a trained graph classification model, and selecting the top R subgraphs that maximize the probability of target class prediction value as the final semantic subgraph; generating a k-order induced graph for each node in the drug molecular graph structure data; setting a similarity threshold, calculating the similarity between the k-order induced graph and the final semantic subgraph, and taking the structure composed of all nodes with similarities greater than the similarity threshold as a key structure; training a multilayer perceptron using the key structure; and adjusting the molecular weight parameters in the process of interpreting the drug molecular graph structure using the trained multilayer perceptron to obtain the optimal interpretability result.

[0013] An efficient parameter fine-tuning method to enhance the understandability of graph neural network explanations, such as Figure 1 As shown, the method includes:

[0014] S1: Obtain drug molecule graph structure data, the topological structure of the drug molecule graph structure data is expressed as G=(V, E, X, A), wherein G represents the drug molecule graph structure; V represents the set of atomic nodes in G; E represents the edge set in G, that is, the set of chemical bonds between atoms in the molecule; X represents the characteristic matrix of atoms in G; A represents the adjacency matrix of G.

[0015] S2: Decompose the drug molecules into multiple subgraphs and embed them into the ordered embedding space. Generate a set of candidate semantic subgraphs through the scoring function iteration with a greedy strategy. Input the set of semantic subgraphs into the graph classification model and select the top R subgraphs that have the largest probability of predicting the target class as the final explained semantic subgraphs.

[0016] G pattern ={g1,…,g R},f(g)=c,g∈G cand .

[0017] Among them, topR represents the semantic subgraphs predicted by GNN as the target category by sorting and selecting the top R probabilities, c is the target category, and f represents the pre-trained GNN.

[0018] Generating a candidate semantic subgraph set includes: classifying and sorting the subgraph set in the embedding space; setting the subgraph size, where the subgraph size is 6 to 8; using a greedy strategy to randomly select nodes from a graph classified as C, and iterate to generate a subgraph; repeating the iteration for 10 rounds to obtain a candidate semantic subgraph set.

[0019] Classifying and sorting the subgraph set of the embedded space includes calculating the scores of the subgraphs of the embedded space and classifying and sorting them according to the scores. The calculation formula is:

[0020]

[0021] Among them, G C Represents a graph classified as C, Z c represents the corresponding embedding.

[0022] The framework structure of the graph classification model is as follows Figure 2 As shown, the expression of the node classification graph neural network model GCN is:

[0023]

[0024] Among them, σ represents the activation function, I represents the identity matrix, represents the node degree matrix, and W represents the parameters that need to be learned in the graph neural network; represents the adjusted new adjacency matrix, A represents the adjacency matrix, and X represents the feature matrix.

[0025] S3: Generate a k-order induced graph for each node in the molecular graph, calculate the similarity between the semantic subgraph and the k-order induced graph, and select the structure composed of nodes with higher similarity as the key structure. The value of k here is generally the diameter of the semantic subgraph.

[0026] The similarity between the k-order induced graph and the final semantic subgraph is calculated as:

[0027]

[0028] Where D is the embedding dimension, Δz d is the difference in d dimensions, represents semantic subgraph embedding, represents the embedding of the induced graph of node u, and ∈ is the similarity threshold.

[0029] By minimizing the distance metric, a set of k-order induced graphs can be obtained to locate similar nodes.

[0030] S4: Train a multi-layer perceptron, the target is the multi-hot encoding of the matched molecular graph, and optimize the multi-layer perceptron by fitting the multi-hot encoding of the original image and the matched molecular graph. The structure of the multi-layer perceptron is as follows Figure 3 As shown, the edge embedding formed by connecting the corresponding embeddings of the nodes using a multilayer perceptron is used as input, and its training formula is:

[0031] Z i,j =W(h i ||h j )+b

[0032] w ij =softmax(Z i,j )

[0033] Among them, w ij represents the weight of the edge, h i and h j represents the embedding of the corresponding node, W is the connection weight, b is the bias coefficient, and Z i,j is the output of the hidden layer.

[0034] S5: Use the trained multi-layer perceptron to adjust the molecular weights of the graph neural network interpretation model during the interpretation process, and use the multi-layer perceptron to adjust:

[0035]

[0036] Among them, M ij Represents the edge mask between node i and node j. The trained multilayer perceptron can flexibly adjust the edge weights so that the weights of matched edges increase and the weights of unmatched edges decrease.

[0037] In this embodiment, the present invention uses an open source pharmaceutical data set Mutagenicity, which contains 4337 compound samples, each of which has a binary label indicating whether it is a mutagenic compound. The Mutagenicity data set is divided into a training set, a validation set, and a test set according to a ratio of 7:2:1, and the data set is trained by GNN. Using the method of the present invention, a batch of semantic subgraphs are generated on the data set classified as a mutagenic compound category by GNN, and the semantic subgraphs are applied to drug molecules for adjustment to obtain new drug molecules. The new drug molecules are input into the trained GNN for verification, and the probability value of being classified as a mutagenic class is very high. The molecular structure of the mutagenic compound category in the Mutagenicity data set is greatly understood, and a reference is provided for designing new molecular compounds of the mutagenic compound category.

[0038] In this embodiment, the present invention uses a public biochemical experimental data set BBBP for drug development research, which contains the amino acid sequence and secondary structure information of 2039 protein molecules. Each sample has a binary label, indicating whether the protein penetrates the blood-brain barrier. The BBBP data set is divided into a training set, a validation set, and a test set according to a ratio of 7:2:1, and the data set is trained by GNN. The method of the present invention is used to generate a batch of new molecules on the data set classified by GNN as being able to penetrate the blood-brain barrier, which can help scientists understand the typical structural characteristics of molecules that penetrate the blood-brain barrier. At the same time, the generated molecules are connected to a certain molecule, which can make the connected molecules classified by GNN as penetrating the blood-brain barrier, helping to design new molecules with the property of penetrating the blood-brain barrier.

[0039] The above embodiments further illustrate the purpose, technical solutions and advantages of the present invention in detail. It should be understood that the above embodiments are only preferred implementation modes of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made to the present invention within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A parameter fine-tuning method for interpretability of drug molecular graph structure, characterized in that: include: Obtain drug molecular graph structure data; Decompose the drug molecule graph structure data into multiple subgraphs, and embed the multiple subgraphs into space; A greedy strategy is used to process the subgraphs in the embedding space, and a set of candidate semantic subgraphs is generated through iterative scoring functions. The candidate semantic subgraph set is input into the trained graph classification model, and the top R subgraphs that maximize the probability of the target class prediction value are selected as the final semantic subgraphs. A k-order induced graph is generated for each node in the drug molecular graph structure data. A similarity threshold is set to calculate the similarity between the k-order induced graph and the final semantic subgraph, and the structure composed of all nodes with similarities greater than the similarity threshold is taken as the key structure. The key structure is used to train a multilayer perceptron. The molecular weight parameters in the process of interpreting the drug molecular graph structure are adjusted using the trained multilayer perceptron to obtain the optimal interpretability result.

2. A parameter fine-tuning method for interpretability of drug molecule graph structure according to claim 1, characterized in that: The topological structure of the drug molecule graph structure data is expressed as G = (V, E, X, A); wherein G represents the drug molecule graph structure; V represents the set of atomic nodes in G; E represents the edge set in G, that is, the set of chemical bonds between atoms in the molecule; X represents the characteristic matrix of atoms in G; and A represents the adjacency matrix of G.

3. A parameter fine-tuning method for interpretability of drug molecular graph structure according to claim 1, characterized in that: The objective function expression of embedding multiple subgraphs into space is: Among them, z q represents the embedding of graph q, z u represents the embedding of graph u, P represents the graph embedding set of positive examples, N represents the graph embedding set of negative examples, α represents the hyperparameter, E(z q ,z u ) is the embedding space.

4. A parameter fine-tuning method for interpretability of drug molecule graph structure according to claim 1, characterized in that: Generating a set of candidate semantic subgraphs includes: classifying and sorting the subgraph set in the embedding space; setting the subgraph size, randomly selecting nodes from the graph classified as C using a greedy strategy, and iterating to generate subgraphs; repeating the iteration for 10 rounds to obtain a set of candidate semantic subgraphs.

5. A parameter fine-tuning method for interpretability of drug molecule graph structure according to claim 1, characterized in that: The graph classification model is the node classification graph neural network model GCN.

6. A parameter fine-tuning method for interpretability of drug molecule graph structure according to claim 1, characterized in that: The similarity between the k-order induced graph and the final semantic subgraph is calculated as: Where D is the embedding dimension, Δz d is the difference in d dimensions, represents semantic subgraph embedding, represents the embedding of the induced graph of node u, and ∈ is the similarity threshold.

7. A parameter fine-tuning method for interpretability of drug molecule graph structure according to claim 1, characterized in that: Training a multilayer perceptron involves using the edge embedding formed by connecting the corresponding embeddings of the nodes using the multilayer perceptron as input. The training formula is: Z i,j =W(h i ||h j )+b w ij =softmax(Z i,j ) Among them, w ij represents the weight of the edge, h i and h j represents the embedding of the corresponding node, W is the connection weight, b is the bias coefficient, and Z i,j is the output of the hidden layer.

8. A parameter fine-tuning method for interpretability of drug molecule graph structure according to claim 1, characterized in that: The molecular weight parameters are adjusted using the trained multilayer perceptron including: Among them, M ij represents the edge mask between node i and node j, w ij is the weight of each edge.

Images