Identification And Prognostic Analysis Of Synthetic Lethal Gene Pairs Based On Decision Tree

Cancer treatment has always been a worldwide problem.In recent years,synthetic lethality has become a new strategy for the development of anticancer drugs.Targeted therapy based on synthetic lethal interactions can effectively overcome the problem of drug resistance.Currently,the main problem of synthetic lethal therapy is the identification of reliable gene pairs in specific cancer type,which can provide data resources for searching potential targets.Due to the higher cost and longer time in experimental validation,bioinformatics approach has become an important method to predict gene pairs with synthetic lethality.In this study,we aimed to predict synthetic lethality using machine learning based on the features of verified gene pairs,in-depthly analyze these predicted data and construct a database.Firstly,sequencing data in 33 cancer types from The Cancer Genome Atlas(TCGA)database were analyzed from different molecular levels,mainly including gene mutation,m RNA,methylation and copy number variation(CNV),and features were extracted to predict synthetic lethal interactions using decision tree model.Secondly,based on predicted results,in-depth analyses were performed to understand the potential biological function and prognostic value.Finally,to share predicted data,we developed a database of synthetic lethality based on web,which would provide a data platform for relevant studies.There were three main parts in this study:(1)Prediction of synthetic lethal gene pairs by using decision tree algorithm:(1)The features in 33 cancer types were firstly analyzed and collected from gene mutation,m RNA,methylation and CNV,respectively,and integrative analysis of these data was performed and quantified.(2)Predicted gene pairs from published papers and databases were used as candidate synthetic lethal gene pairs,and the experimentally validated positive and negative gene pairs were used as train set.(3)The data in(2)were scored,and these specific scores in different cancer types would be features in decision tree.(4)The screened indexes were obtained based on experimentally validated gene pairs through quantifying of values of mutation,m RNA and CNV.(5)Based on the decision tree algorithm,the candidate gene pairs were further screened based on the scores of the relationship between mutation,m RNA and CNV.(6)The model accuracy was evaluated using error rate,precision rate and recall rate,and predicted results were compared with other prediction algorithm.(2)Prognosis and functional analysis of synthetic lethal gene pairs: Based on the predicted synthetic lethal gene pairs,further analyses were performed based on prognosis and biological function.Firstly,the distributions of gene pairs were analyzed,and gene pairs were screened and constructed interaction network if they were shared by at least 10 cancer types.Survival analysis was used to understand the potential prognostic value,the relationships of genes and anti-caner drugs were predicted,and function enrichment analysis was used to understand the potential biological function.Then,take breast cancer(BRCA)as an example to perform further analysis,because BRCA has been widely studied based on synthetic lethality.Gene pairs were screened to perform in-depth analysis from multiple levels to understand the potential roles in occurrence and development of cancer.(3)Construction of online database of synthetic lethality: Based on predicted synthetic lethal interactions,an online database,SLOAD(http://tmliang.cn/SL),was constructed based on web using Django.Except for predicted data,SLOAD also provided visual analysis function,mainly including gene mutation percentage,m RNA expression,Bate-Value of methylation,CNV,and survival analysis based on gene mutation and m RNA expression,which provided a data platform for relevant researchers involved in synthetic lethality.