Prediction Of Pathway Related Protein,Drug And Disease Association Based On Complex Network And Deep Learning

With the development of molecular biology,pathway analysis and the study of disease-pathway relationships have become one of the research hotspots.Biological pathway analysis is an important part of systems biology.Molecular pathways are related to the cellular processes of the entire organism's functions.Abnormal cell behaviors are triggered by defective signal transduction.These abnormal cell behaviors can cause human diseases.Compared with traditional experimental methods,calculation-based methods save time and effort.A variety of calculation methods have been used to study the signal transduction process and the relationship between pathways,diseases,and drugs.However,the existing algorithms do not fully exploit the biological characteristics of the pathway,and the accuracy of the algorithm still needs to be further improved.This article studies the relationship between pathways and diseases,drugs,and proposes new calculation algorithms to predict the relationship between pathways and diseases,drugs.The accuracy of algorithm is significantly improved.At the same time,it combines deep learning and convolutional neural networks,and explores the relationship between pathways and specific protein domains.The main research work is as follows:(1)Describe information related to pathway mechanisms,features,and database feature annotations,discuss the current research status related to data storage and retrieval in biological pathway databases,and analyze the annotations,features,and features of signal pathway databases that can be used to retrieve biological pathway analysis.(2)A predictive model of disease-pathway association RWRH is proposed.First,construct a disease-pathway heterogeneous network,apply a random walk algorithm with restart function to mine seed nodes on the path similarity network and disease similarity network,and then use the PageRank algorithm to rank and score,and then predict potential diseases-Pathway association.(3)A new drug-pathway association prediction calculation method based on known drug-pathway association relationship NCPHDPA is proposed.First,construct a drug-pathway heterogeneous network,integrate cosine similarity and Jaccard similarity to calculate disease similarity and pathway similarity,and then calculate the projection scores of drugs and pathway networks separately to comprehensively predict the relationship between drugs and pathways.(4)A method based on a hybrid feature space and deep learning algorithm to mine the relationship between pathways and specific protein domains is proposed.This method collects protein characteristics such as amino acid composition,chain transfer distribution,dipeptide composition and pseudo-amino acid composition,and constructs a deep neural network to further extract the characteristics.Finally,different classifiers were used to predict the relationship,and the performance of each classifier in the model was analyzed.(5)A method for mining specific protein domains using convolutional neural networks is proposed.Using dipeptide deviation from the expected average value and dipeptide composition information as protein characteristics,a multi-layer convolutional neural network was designed,and different neural network objective function optimization algorithms were compared.The algorithm can fully explore the relationship between pathways and specific protein domains,and the algorithm has a high accuracy rate.