Classification of oral bioavailability of drugs by machine learning approaches: a comparative study

Oral Bioavailability is the rate and extent to which an active drug substance is absorbed and becomes available to the general circulation. A computational model for the prediction of oral bioavailability is a vital initial step in the drug discovery. It is decisive for selecting the promising compounds for the next level optimizations and recognition for the clinical trials. In the present investigation we aimed to perform the oral bioavailability prediction by comparing three machine learning methods i.e. Support Vector Machine (SVM) based kernel learning, Artificial Neural Network (ANN) and Bayesian classification approach. The overall prediction efficiency of SVM based model for the test set was 96.85%, whereas according to the Bayesian classifier and ANN methods prediction efficiency was found to be 92.19% and 94.53% respectively. Thus the present results clearly suggested that the SVM based prediction of oral bioavailability of drugs is more efficient binary classification approach for the data under consideration.