Machine Learning for Classification of Inhibitors of Hepatic Drug Transporters

Abstract

Interactions between drugs may occur when drugs are administered together. These interactions can increase or decrease the efficacy of one of the drugs or can cause a new therapeutic effect which cannot be attributed to either drug alone. An important mechanism underlying drug-drug interactions is inhibition of proteins that mediate transport of drugs across cellular membranes. We developed five machine learning models, including deep learning, for predicting which drugs may inhibit transporter proteins in the liver, and assessed their performance in internal and external validation. Three out of five methods, k-nearest Neighbors, Support Vector Machines, and Recursive Neural Networks have not been previously applied in this domain. The area under the Receiver Operating Curve statistic for the five models ranged between 67% and 78%. Random forest and Support Vector Machines models showed the highest performance in external validation as assessed by the F1 metric. Our modeling approach and results demonstrate a practical application of machine learning techniques in an important application domain.