The Improved Na?ve Bayesian Algorithm For Drug Combination With Pharmacokinetic Consideration

Drug combinatorial therapy is promising for complex diseases than single drug because of their less side effect,lower toxicity and better efficacy.However,the process of discovering effective drug combination is exhaustive.Many of them were found by experimental method which is labor and time consuming.In this study,we used new computational method which combined pharmacokinetic information and improve na?ve Bayesian to predict effective drug combinations.As we know,after drugs entering the human body,it undergoes four processes,absorption,distribution,metabolism and excretion.Considering these processes,this study attempted to use the following features to predict effective drug combination: metabolic enzyme,drug transporter,side effects,targets and KEGG pathway.Compared to previous methods,we added two new features,enzyme and transporter.In this paper,we had constructed the qualitative feature of five types of information,aiming to study the effects of different features on predictive effective drug combinations and the specialty of individual drugs that can be composed of drug combinations.Because of the incompleteness of the information about drug combinations,there were only 17 effective drug combinations that had all five types of information.Therefore,we used drug combination dataset that had five types of features to build the model separately.We found that enzyme had a definitely effect on predicting the effective drug combination and the performance of enzyme was similar to the pathway feature.And in this study we defined two different negative samples and found that individual drugs that had promise to become drug combination with other drugs were more concentrated in existing effective drug combinations.Except this,we used a new feature selection method to select the features and improved na?ve Bayesian algorithm to construct the model.Ultimately,we found that compare to maximum relevance minimum redundancy(mRMR)algorithm,the feature which used our new feature selection method to select was more representative,and the predictive performance was more better than mRMR's.And the performance of improved na?ve Bayesian was better than na?ve Bayesian too.In a word,we defined different negative dataset,used new feature information,new feature selection method and improved Bayesian algorithm in this paper.That was instructive for qualitative understanding of the effects of drug combinations on the body.