Towards Multipurpose Drug Repositioning: Fusion of Multiple Kernels and Partial Equivalence Relations Using GPU-accelerated Metric Learning

Distance Metric Learning ( DML) can be a valuable tool in biomedical data analysis as it can incorporate both partial logical equivalences and multiple similarities, solving the multitask fusion problem by producing a unified view of the entities. However, many implementations are too slow to be applied on real-world data and can benefit from GPU acceleration. Here we present and evaluate a GPU-accelerated DML framework in a drug repositioning context, where DML allows repositioning in a symmetric fashion for both drugs and diseases.