DOCTORAL SEMINAR

Modelling Approaches to Predicting Microbial Interactions

Speaker
Mr Li Chenhao
Advisor
Dr Wong Lim Soon, Kithct Chair Professor, School of Computing
Dr Niranjan Nagarajan, Associate Professor, School of Computing


29 Oct 2018 Monday, 01:30 PM to 03:00 PM

MR1, COM1-03-19

Abstract:

Computational modelling is an important approach to studying the organization and function of "social communities" formed by microorganisms. With the increasing availability of high-throughput microbial profiling data, it is now conceivable to directly learn ecological models that define microbial interactions and explain community dynamics. However, the applicability of these approaches is severely limited by the inability to normalize data in the absence of accurately measured biomass data.

We introduce a novel expectation maximization (EM) like framework that infers both biomass and ecological models from microbial profiling data. Extending our EM-like framework, we developed two novel algorithms (BEEM and BEEM-Static) for learning generalized Lotka-Volterra models from longitudinally and cross-sectionally sampled microbial profiling data respectively. Our benchmarking results demonstrate that BEEM and BEEM-Static outperform state-of-the-art methods for ecological model inference, while simultaneously providing accurate biomass estimation.

By applying BEEM to longitudinal human gut microbial profiles, we were able to uncover personalized ecological interactions and associated dynamics of human gut microbial communities. Moving forward, by mining publicly available datasets with BEEM-Static, we expect to gain insights into human gut microbial ecology on a population scale.