Intelligent Analysis Of Drug-Target Interaction

Drug-target interaction is the basis of new drug development and drug repositioning.Traditional experimental techniques have not met the growing demand for drug research and development due to their high time and money costs.With the development of big data and artificial intelligence in the information age,computer-aided drug design has attracted more and more attention from drug developers.The prediction and analysis of drug-target interactions is an important application of computer technology in drug design.It covers.It covers multiple disciplines and fields,such as genomics,high-throughput screening,molecular dynamics simulation,complex networks,matrix decomposition,and machine learning,and vigorously promotes the development of new drug research and repositioning.This paper focuses on drugtarget interaction,the prediction of drug-target interaction,drug binary classification,and the binding affinity of peptide drugs and receptor proteins.The specific research contents and main contributions are as follows:1.Predicting drug-target interactions using a network-based approach.This section combines multiple drug-target similarities and interaction matrices with the KATZ method to build a model for predicting drug-target interactions,integrating 25 drug similarities and 17 target similarities.They are weighted and integrated with the GAPK similarity again to obtain the drug-target similarity matrix,and then the drug-target similarity matrix and the drug-target interaction matrix are integrated into a heterogeneous network combined with the KATZ method,in the walk,drug-target interactions were predicted with steps of 2,3,and 4,respectively.The results show that the model has the best performance when they walk step size is 2,and the performance of the model decreases with the increase of the walk step size.The model can successfully predict some of the hidden known drug-target interactions and some potential drug-target interactions.Molecular docking experiments verify the credibility of the newly predicted interactions.2.Prediction of drug-target interactions using a matrix factorization-based approach.This part incorporates the drug-target similarity into a probabilistic matrix factorization model.It builds a model named PMF-SDT for drug-target interaction prediction,which integrates the drug’s structural similarity with the target’s sequence similarity drug-target interaction matrix into a heterogeneous matrix.The heterogeneous network is projected into the latent feature matrix of the drug and the target in the network.Then the drug-target interaction matrix is reconstructed as the product of two low-rank matrices,and the drug-target interaction probability matrix is obtained to predict the drug-target interaction.The results show that the model constructed in this part has higher AUC and AUPR values on the gold standard data set and independent test set and better prediction performance for sparse matrices.With the increasing number of drugs and targets,As the drug-target interaction matrix becomes more and more sparse,the model will become more and more useful for predicting drug-target interactions.The model can also predict some hidden known drug-target interactions,and some potential drug-target interactions can be predicted and verified by molecular docking experiments.3.Drug classification using drug-target interaction descriptors.This part takes anticancer drugs and anti-inflammatory drugs as the research objects,obtains drug-target complexes through molecular docking experiments,calculates 67 descriptors of drug-target interaction interfaces,and then selects 22 important descriptions by XGBoost algorithm.As the feature vector,three machine learning models of SVM,Light GBM,and MLP were constructed for drug classification.The results showed that the three models constructed using the drug-target interaction descriptors had higher performances.Compared with the models constructed using the drug small molecule descriptors,the performances of the three models were improved.It also has strong interpretability in biology.Key features are also analyzed in this section.The number of atom pairs,force fields,hydrophobic interactions,and b SASA are key features in identifying anticancer and anti-inflammatory drugs.They work together to influence the classification of drugs.4.Predicting the binding affinity of proteins to peptide drugs using protein-peptide interaction descriptors.This part takes the MHC-I receptor protein and its binding peptides as the research object,obtains protein-peptide complexes through molecular docking experiments,calculates the descriptors of 94 protein-peptide interaction interfaces,and then uses recursive feature elimination algorithm.Ten important descriptors were selected,and three QSAR models for predicting protein-peptide binding affinity,SVR,RF,and MLP,were constructed.The results showed that compared with the peptide descriptors,the three models constructed using the protein-peptide interaction descriptors exhibited lower errors and higher coefficients of determination and had stronger predictability in pharmacy and biology explanatory.The key features are also analyzed in this section.During the binding process,the b SASA of negatively charged species,hydrogen bond acceptors,hydrophobic groups,planar structures,and aromatic rings of peptides are key features for predicting protein-peptide binding affinity.The binding affinity of proteins and peptides is determined under the joint action.To predict the binding affinity of other proteins and peptides,the research content in this section can also be used as an important reference.All in all,this paper focuses on drug-target interaction,constructs research models suitable for different scenarios from multiple perspectives,and provides an important reference for drug treatment of diseases.In the era of big data,the research content of this paper can provide new ideas for precision medicine.