AF: Small: Homogeneous and Heterogeneous Network Learning with Applications in Computational Biology

Case Western Reserve University, Cleveland OH

Investigators

Abstract

Heterogeneous networks have been widely used in modeling real-world complex systems and have been a powerful tool in studying complex biological problems. Link prediction in heterogeneous networks has been one of the key computational problems in this context. However, research on efficient and effective algorithms for link prediction in heterogeneous networks is still in its infancy, especially for networks that integrate multiple data sources. The overall objective of this project is to develop efficient, robust, effective and integrative algorithms as well as software tools for the link prediction problem in heterogeneous networks, and to apply the algorithms on real biological applications with practical significance. The project will also help in promoting teaching and training of a new generation of computational biologists. The investigators of the project will solve the link prediction problem in heterogeneous networks by proposing two computational approaches: one is based on a kernel regularized least square framework with multi-view learning for data incompleteness, and the other one is to utilize a weighted nonnegative matrix tri-factorization approach that provides a unified framework for link prediction and data imputation. Both approaches will be applied to the problem of drug combination prediction, which is formulated as a link prediction problem in heterogeneous networks. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.