LINGOBLM: using LINGO kernel in Bipartite Local Model

Abstract

Predicting potential drug-target interactions from heterogeneous biological data could benefit novel drugs discovery and improve human medicine. Computational prediction is a suitable alternative for the traditional time-consuming and expensive experimental process of drug-target interactions prediction. New computational drug-target interactions prediction approaches are divided into two categories: machine learning-based and network-based. In this paper, we extended the Bipartite Local Model (BLM), one of the most well-known approaches for predicting drug-target interactions. BLM has a high computational complexity due to the use of a two-dimensional kernel for the drug side. Instead, we used LINGO, a one-dimensional kernel, to calculate the similarity between drugs. In order to compare our work with previously published results, we performed experiments using publicly available real-world drug-target interactions datasets. The results suggested that our approach is competitive and outperformed BLM.