Prediction Of Drug-induced Hepatotoxicity Based On Machine Learning

Drug-induced hepatotoxicity may cause acute and chronic liver disease,causing much concern for patient safety.On the other hand,this is one of the main reasons for the withdrawal of drugs from the market.To make sure the hepatotoxicity of the drug,researchers used to use animal experiments to predict drug toxicity.But this method is very inefficient,So,drug toxicity genomics data has been widely used in liver toxicity prediction.In our study,we proposed a multi-dose calculation model based on gene expression and toxicity data to predict drug-induced hepatotoxicity.In this paper,based on the dose-response curve,the dose/concentration information after drug treatment was fully utilized,so a more abundant dose-response relationship was considered.Another innovation of our work is proposing a new feature selection method called MEMO,which is also an important aspect of the multi-dose model in our work for processing high-dimensional toxicology genomics data.We validated the proposed model using TG-GATE,a large database that records toxic genomics data from multiple perspectives.Through MEMO algorithm,the gene data was reduced from 9998 to 268,and the accuracy of the model increased from 53.7% to97.1%.The experimental results show that drug-induced hepatotoxicity can be predicted with high accuracy and efficiency using the proposed prediction model.