Prediction Of Compound Toxicity Based On Ensemble Of Classifiers

The toxicity and mutagenicity of compounds are one of the main reasons for the decline in the number of new drugs in the pharmaceutical industry.Researchers currently use a variety of methods to assess thise risks,including in vivo,in vitro,and computer simulation methods.The study found that quantitative structure-activity relationship(QSAR)and other calculation methods have the advantages of obtaining fast results and no need for drug experiments in safety screening in the early stages of drug discovery.However,the prediction performance of most published human drug-induced liver injury(DILI)and compound mutagenesis models is not satisfactory,and the accuracy needs to be improved.In addition,limited reliable data,such as human DILI and bacterial reverse mutation test(Ames test)are,is also a huge obstacle to computational modeling.This article dissertation mainly conducts related research from the aspects of toxicity and mutagenicity of the compounds:1.The work Propose a drug-induced liver injury prediction based on integrated classifier method.The development of new drugs is limited by multiple factors.Even if many drugs can reach the stage of clinical trials,drug development will still fail due to factors such as druginduced liver injury.The National Toxicology Research Center(NCTR)of the US Food and Drug Administration has established a data set on drug-induced liver injury(DILI),and established a good predictive model based on this data set to provide a reference for future drug development.But the model is based on a single classifier model for prediction,and the accuracy of prediction is very low.In this paper,based on the NCTR publicly published data set,a multi-base integrated classifier-based method is used for modeling.From the 12 fingerprints,8 fingerprints are selected that have a relatively high correlation with the prediction of drug-induced liver injury.Seven molecular descriptors were generated,and the optimal ratio of fingerprints and molecular descriptors to the model was tested.From the nine base classifiers,the five best classifiers were found,and the five base classifiers were modeled by means of Voting integration,and achieved good results in predicting drug-induced liver injury.2.The work research on compound mutagenicity prediction based on integrated classifier method.This article work uses the published Ames mutagenicity benchmark data set to model and predict whether related compounds will cause mutations,so as to carry out compound mutagenicity studies.First of all,based on the published data set,we draw on the relevant work experience of the prediction of drug-induced liver injury,and select 5 fingerprint maps with a large contribution to the model from 12 fingerprint maps.Choose the five base classifiers with the best performance from the nine base classifiers,and perform integrated modeling based on the classic integration method Stacking,and obtain good prediction results.This research can better provide a reference for the development of drug research,greatly reduce the huge waste of human,material and financial resources brought by traditional drug research and development,and provide guarantee for the development of new drugs.