Large-scale Analysis of Drug Combinations by Integrating Multiple Heterogeneous Information Networks

Personalized treatments and targeted therapies are the most promising approaches to treat complex human diseases. However, drug resistance is often acquired after treatments. To reduce drug resistance, combinational drug therapies have been considered as a promising strategy in drug discovery. Moreover, the emerging of large-scale genomic, chemical and biomedical data provides new opportunities for drug combinations. In this work, we propose a network approach, called MCDC, that integrates multiple data sources describing drugs, target proteins, and diseases to predict beneficial drug combination. Specifically, MCDC integrates diverse drug-related information (e.g., chemical structure, target profile), disease-related information (e.g., disease phynotype), together with their interactions to construct a two-layer heterogeneous network. MCDC then predicts drug combinations for each disease using a link prediction algorithm. Due to the nature of data collection, missing data is common in systematic integration of these heterogeneous data. We further develop a multiple incomplete view learning algorithm to address the issue of missing data. Extensive experimental studies show that the proposed method outperforms several network-based methods. Our approach has great potential to accelerate the development of drug combination treatments.