Mining Multi-Layer Protein-Protein Association Networks: An Integrated Spectral Approach

Tufts University, Medford MA

Investigators

Abstract

This project is focused on strategies which help us to obtain information for associated pairs of proteins or genes in biological systems. In addition to the collection of lots of information about the role of different genes or proteins in the cell, there is also increasing information about pairs of proteins or genes that are related, either because there is evidence that they cooperate in the cell, or evidence that they otherwise have common attributes. The information about associated pairs can be mathematically described as a heterogeneous collection of networks, but designing efficient and effective machine learning and computational mathematics algorithms to integrate the diverse information sources to explore and make sense of these networks is a difficult unsolved problem. The new mathematical methods that will be developed in this project will be customized for the computational biologists and systems biologists who would like to use network analysis to boost the statistical significance of the signal of important genes and pathways in their data, with applications to gene function prediction, and the identification of sets of genes that are important in complex diseases such as type II diabetes and Crohn's disease. The project supports one graduate student and two undergraduate students. Through training and collaborating with investigators and other experts in the field, they will become involved in the broader research communities of scientific computing and biology. Effective and efficient inference and computational methods will be developed, analyzed, and implemented for mining multi-layer PPI networks via an integrated spectral approach and the generalizations of diffusion-based distance metrics. More precisely, spectral multilayer analysis methods based on dimension reduction and multilevel optimization methods will be designed in order to provide high-quality integration tools of multiple networks that can be used to mine this massive graph collection. The methods will be benchmarked and tested on a substantial new biological network testbed, connected with the recent DREAM disease module identification challenge. Furthermore, implementations of the tools will be made generally available to the community, for mining heterogeneous network collections in general, which will lead to new insights related to core problems in computational biology, including the identification of disease modules within the datasets. 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.