Title: Computational methods for analyzing variation in the microbiome
This event is hosted as part of the seminar series featuring leaders in precision medicine from across the nation. Co-presented by the Center for Precision Medicine and Genomics, the Columbia Precision Medicine Initiative, and the Precision Medicine Resource of the Irving Institute for Clinical and Translational Research. This event is supported in part by a gift from Pfizer to Columbia.
Speaker: Itsik Pe'er, Professor and Vice Chair, Department of Computer Science, Fu Foundation School of Engineering and Applied Science, Columbia University.
Location: Zoom and in-person. Register to receive the zoom link for this event.
Description: The advent of longitudinal microbiome data has highlighted the relationship between dynamic changes in resident microbial composition and disease. Yet, analysis of such data is hindered by technical noise, high dimensionality, data sparsity, and the measurement of relative, rather than absolute abundance. We introduce a battery of methods for inferring dynamics of longitudinal microbiome compositions, * LUMINATE is a method to infer abundances from noisy read count data. It employs an explicit distinction between biological zeros - null read counts due to the absence of a taxon - and technical zeros - due to sampling issues. Technically, it uses variational inference to be orders of magnitude faster than current approaches, with better or similar accuracy, making it feasible to analyze large datasets. * Lotka-Volterra (predator-prey) differential equations have been used to model the dynamics of microbial species. We generalize this framework to handle compositional data of relative abundances. * We employ machine learning methods to evaluate peak-to-trough ratios of read coverages at the replication origins vs. termini and infer growth rates, expanding the scenarios in which this analysis can be executed.
