Predictive and Interpretable Bayesian Machine Learning Models for Understanding Microbiome Dynamics

• Activity Number: 48 - Longitudinal Modeling and Experimental Design for Investigating Host Associated Microbiota
• Type: Invited
• Date/Time: Sunday, July 29, 2018 : 4:00 PM to 5:50 PM
• Sponsor: IMS
• Author(s): Georg Kurt Gerber*
• Companies: Harvard Medical School / Brigham and Women's Hospital
• Keywords: Bayesian; time-series; dynamical systems; microbiome; computational biology; interpretable models

Abstract

The human microbiome is highly dynamic on multiple timescales, changing dramatically during development of the gut in childhood, with diet, or due to medical interventions. I will present two novel Bayesian machine learning methods for gaining insight into microbiome dynamics. The first, MDSINE, is a method for efficiently inferring dynamical systems models from microbiome time-series data. I will present recent extensions to MDSINE including interaction modules, or learned clusters of latent variables (reducing the expected number of interaction coefficients from O(n^2) to O((log n)^2)); a fully Bayesian stochastic dynamical systems formulation that propagates measurement and latent state uncertainty throughout the model; and a temporally varying auxiliary variable technique to enable efficient inference by relaxing hard non-negativity constraints on states. The second method, Microbiome Interpretable Temporal Rule Engine (MITRE), is a method for predicting host status from microbiome time-series data, which achieves high accuracy while maintaining interpretability by learning predictive rules over automatically inferred time-periods and phylogenetically related microbes.