Learning microbial dynamics for therapeutic applications: Scalable inference, robustness, and control.
, Research Fellow Harvard Medical School & Brigham and Women's Hospital
Date: Wednesday, April 10, 2019
Time: 11:30 AM to 1:00 PM
Location: Stata Center Building 32 Room G575
Event Type: Seminar
Host: Bonnie Berger, CSAIL and Mathematics
Contact: Patrice Macaluso, 617-253-3037, firstname.lastname@example.org
Speaker URL: None
email@example.com, firstname.lastname@example.org, email@example.com
TALK: Learning microbial dynamics for therapeutic applications: Scalable inference, robustness, and control.
Microbes are everywhere, including in and on our bodies, and have been shown to play key roles in a variety of prevalent human diseases. Consequently, there has been intense interest in the design of bacteriotherapies or "bugs as drugs," which are communities of bacteria administered to patients for specific therapeutic applications. Central to the design of such therapeutics is an understanding of the causal microbial interaction network and the population dynamics of the organisms. Toward that direction I will present recent work on a Bayesian nonparametric model and associated efficient inference algorithm that addresses the key conceptual and practical challenges of learning microbial dynamics from time series microbe abundance data (Ref 1). These challenges include high-dimensional (300+ strains of bacteria in the gut) but temporally sparse and non-uniformly sampled data; high measurement noise; and, nonlinear and physically non-negative dynamics. In a related work I will discuss a simpler inference problem surrounding the engineering of an interdependent consortia of bacteria (Ref 2). Here we will focus on experimental design for inference and discuss best practices for designing synthetic bacterial consortia for clinical/pharmaceutical applications.
If time is left, I will also discuss recent work on analyzing gradient descent where we provide provably stable accelerated algorithms for optimization when features are time varying (Ref 3). If we wish to deploy learning algorithms in a clinical setting, then we must have provable convergence guarantees and robustness properties similar to the guarantees we have for control algorithms.
Bio: Travis just joined the faculty of Harvard Medical School in the Department of Pathology at Brigham and Women’s Hospital (Division of Computational Pathology). He did post-doctoral training in statistical inference and experimental biology with Dr. Georg Gerber and in the area of microbial dynamics/networks with Dr. Yang-Yu Liu. His PhD was in Control Theory with a minor in Mathematics (Analysis) from MIT. Outside of academic research Travis has worked for NASA, Boeing, and Johnson and Johnson. Travis’ algorithms are currently flying on air vehicles (unmanned and experimental aircraft, NOT the Boeing 737 Max 8).
Created by Patrice Macaluso at Thursday, March 21, 2019 at 12:08 PM.