Prediction Of Protein-protein Interaction By Ensemble Neural Network And New Coding Method

In recent years,with the development of computer technology and the explosive growth of biological macromolecular data,the trend of applying mathematical theory and computer algorithm to biological research is more and more obvious.More and more scholars use machine learning algorithm to predict protein-protein interactions.Under this background,this paper studies the application of neural network,an important branch of machine learning,in the prediction of protein-protein interaction,and proposes a new model,IT-Ens BPNN,to predict protein-protein interactions.The experimental results of this paper show that the averaged accuracy of IT-Ens BPNN is 93.64% in Saccharomyces cerevisiae data set and 97.55% in human data set.Firstly,the first chapter introduces the background and history of protein-protein interaction,points out the significance of using machine learning method to study proteinprotein interaction,and analyzes the research status of protein-protein interaction.In the second chapter,we introduce the first step of the prediction model,which is data preprocessing.In this paper,we first propose a new coding method: Integral Transform encoder,which transforms the original protein amino acid sequence into equal length digital sequence.Then the high-dimensional data is transformed into the low dimensional data that the classifier can run through the data dimension reduction algorithm.Here,this paper introduces the factor analysis method based on the expectation maximization algorithm in detail,and gives two theorems and proofs as the explanation of the key process of the factor analysis method.The third chapter of this paper gives the second step of prediction model,which is also the key of protein-protein interaction prediction model,namely the construction of ensemble neural network classifier.In this chapter,we first introduce the basic knowledge of artificial neural network,and write the composition unit,loss function and optimization algorithm of neural network into a unified mathematical format.Then we introduce the integration method based on the average of random weights and the result integration method based on the degree of uncertainty,so as to form the integrated neural network model we need.The last chapter and conclusion focuses on the prediction results of the prediction model on the data set,and based on these results,a detailed analysis of the performance of the prediction model proposed in this paper is carried out.In this part,we first introduce the various indicators and graph curves used to evaluate the performance of the prediction model,and then intuitively show the excellent performance of the encoder in this model compared with the traditional coding method in the form of graph,as well as the advantages of the classifier in this model compared with other classical classifiers.At the end of the paper,the innovation,deficiency and future research direction of this paper are given.