Computation of models for prediction of blood brain barrier permeability using molecular descriptors

Abstract

Models for prediction of Blood Brain Barrier permeability were developed in the present study. A dataset of 63 structurally diverse compounds was selected for the present investigation. The values of 21 Constitutional descriptors, Topological descriptor and Information indices for each compound of the data set were computed using freely available e-Dragon software. Data generated was analyzed and suitable models were developed using decision tree, random forest and moving average analysis. The accuracy of the models was assessed by calculating overall accuracy of prediction, sensitivity, specificity and Mathew's correlation coefficient. Random forest correctly classified the analogues into permeable and impermeable with an accuracy of 92.06%. A decision tree was also employed for determining the importance of various molecular descriptors. The decision tree learned the information from the input data with an accuracy of 96.82% and correctly predicted the cross-validated (10 fold) data with accuracy up to 84.12%. Accuracy of prediction of proposed moving average analysis models was found to be 96.22% and 90.09%.