Research On Prediction Of Compound And Protein Binding Based On Densely Connected Neural Network Model

Traditional drug development(the method of synthesizing new substances)is time-consuming,expensive,and has a low success rate.This model has encountered bottlenecks in development.The development of modern experimental chemistry,theoretical chemistry,pharmacology,and toxicology has revealed the characteristics of binding between some compounds and proteins,but it is still limited,which restricts the process of drug development.The method based on natural plant,animal,and mineral chemical components has always occupied an important position in the drug development process,and has accumulated a large amount of experimental data.How to extract more characteristics of binding between compounds and proteins from these experimental data,and predict the binding relationship between new compounds and proteins,has important implications for drug development.In recent years,with the rapid development of artificial intelligence,experimental methods based on deep learning have been actively introduced in various fields,and breakthrough progress has been made.Deep learning is also called layered learning.This method is mainly a model inspired by biological brains.Deep learning automatically extracts features through layer-by-layer changes on a large scale of raw data,and automatically learns the extracted hierarchical features.Since deep learning can automatically extract features,there is no high demand for the researcher’s professional knowledge,and it is fast and accurate.It has a great advantage for experiments that process large amounts of data,so as to extract more compounds and proteins The binding feature provides a new way.This paper uses the dense connections in deep learning models,strengthens the model’s information exchange by adding jump connections between different layers,helps the model to learn more robust features,and can also solve the problem of gradient disappearance in deep neural network training,Guaranteed model convergence.The deep neural network model finally used in this paper consists of compound feature networks,protein feature networks,and joint discriminative networks.Each network consists of 3,3,and 4 densely connected blocks.The number of neurons in each densely connected block is 1945,1912,1284,1386,2107,1027,2652,1515,2345,2325,the total number of parameters is about 36.63 M.In this paper,the final model is trained from data processing,construction,training,and optimization.This process tries multiple methods and takes two years.The optimal model has an Accuracy of 98.29%,a Precision of 93.55%,and the F1-Score is 94.04% and the Recall is 94.54%.The research on the binding relationship between compounds and proteins based on the densely connected neural network model used in this paper has obtained good results,providing new ideas and experimental data for subsequent research.