Presenter: Michelle Badri, Ph.D. Candidate New York Univesity
Title: Applying natural language processing methods to building models of the microbiome
Integrating datasets from multiple “omic” datasets is crucial to microbiome research. However, inferring interactions across multiple data types can present many statistical challenges. Our approach to solving this problem for microbe-metabolite association is using neural networks (adapted from the field of natural language processing) to estimate the conditional probability that each molecule is present given the presence of a specific microbe. I will discuss several applications of our microbe-metabolite association learning tool (mmvec) on other multi-omic microbiome studies. Additionally, this design can be readily adapted to learn microbe-microbe interactions. We use publicly available mouse gut microbiome datasets to show that our learned interactions have similar changes in abundance in response to experimental treatments.