Learning Multi-relations in Biomedical Networks and Tensors
Thursday, November 8, 2018 - 11:40am - 12:10pm
Biomedical knowledge graphs represent relations among biomedical entities and have been intensively analyzed for drug repositioning, disease gene discovery and other important medical applications. While most existing studies focus on analyzing and predicting pairwise relations, we consider learning multi-relations among the entities across many large biomedical knowledge graphs. We introduce a tensor-based framework of applying label propagation on the tensor product of multiple graphs for multi-relational learning. We propose an optimization formulation and a scalable Low-rank Tensor-based Label Propagation algorithm (LowrankTLP). The optimization formulation minimizes the rank-k approximation error for computing the closed-form solution of label propagation on a tensor product graph with efficient tensor computations in LowrankTLP. Acceleration with parallel tensor computation enabled label propagation on a tensor product of a large number of big graphs to predict author-paper-venue in publication records, alignment of protein-protein interaction networks across species and alignment of segmented regions across many CT-scan images. We will also present the preliminary results of applying a similar tensor approach to drug-disease-gene association prediction for pharmacogenomics.