Bayesian Ranking Prediction for Drug-Target Interactions

This page contains supplementary information for the paper "Drug-Target Interaction Prediction: a Bayesian Ranking Approach" by Ladislav Peska and Krisztian Buza.


In silico prediction of drug-target interactions (DTI) could provide valuable information and speed-up the process of drug repositioning – finding novel usage for existing drugs. There are two gen-eral drug-repositioning scenarios, namely drug-centric and disease-centric approaches. Drug-centric approach aims to find novel usage for existing or abandoned drugs, while disease-centric approach focus on applying existing drugs on specific (often rare or neglected) diseases. However, the DTI prediction studies presented so far, did not respect these scenarios in their evaluation pro-tocols and, often, in the underlying models either.
We propose a novel matrix factorization method Bayesian Ranking Prediction of Drug-Target Interactions (BRDTI). The method is based on substantially extended Bayesian Personalized Ranking matrix factorization (BPR). The key feature of BPR is that it is optimized to rank potentially interacting targets for each drug (or vice versa) separately instead of learning a global rating of all possible interactions. Such approach is beneficial for research targeted on a specific drug (or tar-get). Evaluation on four benchmark datasets shows that BRDTI can outperform several state-of-the-art approaches in terms of per-drug and per-target nDCG while remaining competitive in more commonly used AUC and AUPR. Furthermore, BRDTI was shown to predict well also novel drug interactions not contained in the original datasets.

Available resources: