Drug Repositioning by Bipartite Local Models

Computational drug repositioning is a promising approach to reduce the costs of investing on new drug indication discovering for drug development. Many computational drug repositioning strategies have been proposed which use global models for all drugs/diseases in predicting new drug-disease associations. However, using a global model to predict novel drug-disease associations for each drug/disease may be less effective because drugs/diseases have different pharmacological and biological features. Using multiple local models for each drug/disease, therefore, could be a potential approach to overcome this limitation. In this study, we present a novel method named as BLMDR that use local models for both drugs and diseases presented in a bipartite network for inferring drug-disease associations. The prediction of a potential association between a pair of drug and disease is made based on two local prediction models, one for the drug and one for the disease. Experiment results on benchmark datasets show that the BLMDR achieves reasonable prediction performance and outperforms a state-of-the-art global model-based prediction method. Furthermore, top predictions of drug-disease pairs supported with evidence from literature could prove the potential of our method in discovering new drug indications.