Prediction Of Drug-drug Interactions Based On Pharmacological Attributes And Drug Multi-omics Information

Drug-Drug Interactions(DDIs)refers to the phenomenon that the original pharmacodynamics,pharmacokinetics or physicochemical properties of drugs change when one or more drugs act on the human body at the same time.According to the different changes,it can produce beneficial effects such as increased efficacy,reduced dosage,and reduced toxic and side effects,or adverse effects such as decreased efficacy and increased toxic and side effects.Drug interactions play an important role in drug research.Traditional drug interaction research is based on statistical data and clinical trials,which is costly,time-consuming and labor-intensive,and has certain risks.With the development of computer technology,machine learning technology has been widely used in the field of drug interaction research.The existing machine learning methods can generally predict DDIs more accurately,but there are still many problems.For example,the research of document extraction method is limited to the existing text data;The method based on similarity is difficult to deal with the relationship between the drug itself and the organism;The method based on link prediction ignores the influence of physiological environment on DDIs.In view of the above problems,this paper proposes a scheme to predict DDIs by simulating the complex physiological environment of drugs and human body based on pharmacological properties and drug-related multi-group information.The specific scheme is as follows:1)The similarity model is built based on the physical and chemical properties of drugs,such as drug molecular structure,pharmacokinetic parameters,etc.The physical and chemical properties data of drugs are represented as the feature vectors of drugs,and the Jaccard similarity measure is used to calculate the similarity between two drugs,and the similarity matrix is constructed as the feature embedding of the physical and chemical properties of drugs.2)Based on the multi-omics data of drugs,such as drug genomics,drug metabolomics,etc.,construct the heterogeneous graph of drug multi-omics information,use the heterogeneous graph attention network to extract the topological structure information and semantics,as the embedded representation of drugs in the multi-omics environment.3)Integrate the results in the first two parts as the final embedded representation of drugs,input them in pairs into the deep neural network for learning,and use the softmax function for multi-classification to predict the potential interaction types between drugs.From the perspective of the integrity of drug interactions,this topic simulates the effects of drugs in vitro and in vivo,and predicts the interactions between drugs after integrating the results,so as to improve the accuracy and interpretability of the prediction of the interactions.This study will provide a method for the study of drug interactions and a new approach for the study of clinical drug combinations.