Prediction Of Drug Target Interactions Based On Deep Learning

Drug therapy is an important means of curing diseases,and the identification of drugs and target proteins is the key to the development of new drugs.However,due to the limitations of high throughput,low precision and high cost of biological experiment methods,it has a certain degree of blindness to verify the interaction of a large number of drug targets,making it difficult to be widely carried out in practical applications.Driven by information science,intelligent information processing technologies have been rapidly developed and applied.By using computer simulation to predict the interaction between drug-target,it can shorten the time,reduce the cost of research and development,and reduce the blindness of new drug development.This article predicts drug-target interactions based on amino acid sequence methods,and proposes numerical characterization methods of drug compound molecules and protein amino acid sequences,feature extraction methods based on protein amino acid sequence information,and powerful deep learning models for drugtarget interaction Make classification predictions.First,the drug and protein information are stored in the biological information database in complex characters,it cannot be directly input into the classifier as feature vectors.This paper proposes a numerical characterization method based on molecular fingerprint characteristics of drug compounds and based on score specificity matrix(PSSM matrix)to achieve quantitative description of target data.Secondly,there is the problem of noisy data and large dimensions based on the numerical features of protein sequences.In this paper,the Variational Auto-Encoder is used to extract features,generate key features with high efficiency and without losing biological information,reduce feature dimensions,remove the influence of noise data,and provide important data guarantee for subsequent classification prediction.Finally,for the construction of drug target prediction models,this paper proposes to use a powerful deep learning classification model to predict drug-target interactions,to achieve efficient prediction of multiple samples and high-dimensional data,and to improve network training speed and drug target prediction accuracy.