Drug-Target Binding Affinity Prediction Based On Power Graph And Word2Vec

Drugs and protein targets affect physiological functions and metabolic effects through bonding reactions,and accurate prediction of the interaction between drugs and protein targets can provide strong support for drug development.In order to reduce drug development time and labor costs,machine learning methods are gradually playing an important role in the field of drug-target interaction.Most drug-target interaction prediction models based on machine learning are based on binary classification methods,but drug target affinity based on regression method can better represent the strength of binding ability,accurate prediction of drug target affinity can effectively reduce the time and cost of drug redirection and new drug development,and affinity is also the key information to judge drug dosage and drug side effects.In this paper,a drug target affinity prediction model based on Graph Neural Network and Word2vec(WGraph DTA)is first proposed.The model starts from the onedimensional representation of the drug and the target protein: for the drug molecule,the one-dimensional sequence SMILES,which represents the molecular structure,is first converted into a molecular map,and then the graph neural network is used to extract the structural information of the drug molecule;For target proteins,the n-gram algorithm is used to divide the amino acid sequence into sentences containing "biological words",which contain the context information of the amino acid sequence,and then use the pretrained Word2vec(Word to vector)dictionary to convert the sentences into an embedding matrix,and extract the features of the target protein through a three-layer convolutional neural network.Finally,the extracted drug and target features are fused and input into the three fully connected layers to obtain the predicted value of drug-target affinity.The experimental results show that the model has a certain improvement in the prediction of drug-target affinity.In order to further improve the performance of the model,this paper introduces the idea of power map,that is,a one-hop neighborhood embedding matrix that can obtain a drug molecular map through neighborhood representation.After removing the one-hop neighborhood node and the central node itself,an embedding matrix with a hop count of two can be obtained,also known as a quadratic power graph.Similarly,a cubic power plot can be obtained by hopping three times.Based on the synthesis of one and twocubic power graphs,the drug target affinity prediction model based on Power graph and Word2vec(WPGraphDTA)was proposed by using graph neural networks to extract drug molecular features,and combining the protein features obtained by word embeddings.On the two benchmark datasets,the prediction results of this model are further improved,showing good prediction performance.