Drug Repositioning Predictions by Non-Negative Matrix Tri-Factorization of Integrated Association Data

Drugs repurposing (i.e., the reuse of existing drugs for new medicalindications) is attracting the interest of pharmaceutical companies,as it speeds up the drug development process and substantially re-duces the need for clinical trials. Thus, computational methods fordrug repositioning are gaining increasing interest. In this work, wepropose a drug repositioning algorithm based on the Non-NegativeMatrix Tri-Factorization (NMTF) of integrated association data. Weshow how to construct a general-purpose graph encompassing themost relevant aspects in drug discovery and how to ensure fast con-vergence of the algorithm. In particular, we study how initializationand termination may significantly affect the outcome quality for thedrug repurposing application. We also evaluate our computationallypredicted repurposed drugs based on the literature and find confir-mation for our prediction, proving that our method can successfullybe applied to hypothesis generation for drug repurposing.