Machine Learning Methods for Modeling Structure in Single-cell Chromatin Data

Abstract

ABSTRACT:

In recent years, single-cell technologies have revolutionized biology, providing exciting opportunities to map cellular populations through development. Among these technologies, single- cell ATAC-seq has become the leading assay for probing the cellular regulatory landscape by mapping chromatin information. How do we discover structure in this high-dimensional data and use it to understand development? Here, I will present a Bayesian state-space model to characterize chromatin information by modeling the duration of functional and accessible chromatin regions, termed ChromA. I will introduce hidden semi-Markov models as a biologically plausible assumption to distill regulatory regions from ATAC-seq data sets. Next, I will show how this model can be extended to analyze single-cell ATAC-seq information and to compare different cellular populations. Finally, I will show how ChromA can be used to map the chromatin developmental landscape of Interneurons in the cerebral cortex, revealing the fundamental logic of cortical interneuron specification.

BIO: Mariano Gabitto works in the fields of machine learning and computational neuroscience. His research focuses on applying statistical and machine learning methods to decipher how networks of genes and molecules within cells interact to give rise to the cellular diversity observed in organisms. He completed his Ph.D. in Neuroscience at Columbia University working with Charles Zuker. While there, He collaborated with Liam Paninski, Larry Abbott and Tom Jesell. After completing his Ph.D., Mariano visited Mike Jordan's Group at U.C. Berkeley to work on Bayesian nonparametric methods for superresolution imaging. Currently, He is a research scientist at the Flatiron Institute's Center for Computational Biology, working with Richard Bonneau (F.I. / New York Univ.) and Gordon Fishell (Harvard Medical School / Broad Institute).