Research On The Prediction Of Drug-target Interaction Based On Probabilistic Relation Autoencoder

The design and development of drugs is based on the knowledge of biological targets and molecular activity to find effective drugs.Drugs are mostly organic molecules or compounds that have a special effect on a variety of proteins called targets by activating or inhibiting the function of biomolecules to product a therapeutic effect on the disease.The interaction between drug and target is important in drug research,such as promoting drug discovery processes,drug side-effect prediction,and drug re-use.The original research method mainly used clinical biological experimental methods to ensure the validity of the experimental results.In the experiment,it was a challenging and expensive process to find a compound with high affinity for the target.This requires the development of more efficient and efficient biometric methods to predict drug-target interactions.Biometric methods are able to mine potential features from the vast amount of data.At the same time,the prediction results of the calculation method can make the biological experiment more targeted and cost-effective.The current calculation methods for predicting the interaction of drug-target are mainly divided into three categories:(1)ligand-based methods;(2)target-based methods;and(3)machine learning methods.Ligand-based methods are commonly used to design a drug based on the relationship between the chemical structure and activity of the ligand when the 3D structure information of the target is unknown.The target-based methods requires the 3D structure of the target.This method simulates the spatial size,shape,and binding pattern of drug and target binding to predict the binding affinity between the drug and the target.Machine learning methods,especially deep learning,have achieved good results in various fields.Researchers have used machine learning to propose many innovative ideas on drug-target relationship prediction.Based on the above description,this work proposes a method based on probabilistic graphical model,and designs a Variational Auto-Encoder to predict the interaction of drug-target.According to the characteristics of the data set,the experiments are mainly divided into two categories: binding relationship prediction and binding affinity prediction.We did a different experimental comparison on several commonly used databases.The results show that our method has a good performance in each experiment,and even in some experiments to achieve the best results.In addition,by the characteristics of probabilistic graphical model,we can also extend our method to infer the drug which can bind for a given protein directly.