Presenter: Michelle Badri, Ph.D. Candidate, New York University
Title: Instrumental variables design for high-dimensional microbiome data
Abstract: Instrumental variables estimation is an approach to causal inference commonly used in the field of econometrics. If X is an explanatory variable and Y is an outcome of interest we hope to establish a causal relationship independently of confounding variables. A confounding variable can potentially have a causal effect on the outcome of interest and/or can be associated with a predictor of interest. In this causal framework we select or design another variable: Z, which is associated with X but not associated with Y except through its association with X. We call Z an instrumental variable that allows us to avoid issues of measurement error, confounding, selection bias, and reverse causality. To satisfy dimensionality constraints in our implementation of instrumental variables analysis we use rank-reduced regression as the first stage in a two-stage least squares regression. We apply this instrumental variables model to various studies in the microbiome, using experimental treatment like transfer of microbes, antibiotic treatment and other controlled perturbations as an instrumental variable to estimate the effect of the microbiome on disease.
Upcoming Presenters:
4/26/21: Meet Barot
5/3/21: Daniel Berenberg
