Integrative characterization of cell state via modeling of multi-omics data

PROJECT SUMMARY Advent of singe-cell genomics has enhanced our ability to study heterogeneous cell populations (1) to track the time course of cellular differentiation and identify drivers, (2) to identify potentially novel cell states/types in an unsupervised manner, and (3) to identify cell populations that are linked disease outcomes. More recently, we and others have developed single-cell multiomics technologies that enable measuring of multiple modalities of a single cell at the same time, including the transcriptome, the methylome, the epigenome and surface marker proteins. These assays offer unprecedented opportunities to study the state of single cells more comprehensively; by developing the necessary computational methods for these assays, we can obtain more accurate and deeper characterization of cell states and obtain mechanistic insights into the relationships between state of chromatin, the proteome and the transcriptomic states within an individual cell. However, there is a dramatic lack of interpretable computational methods to study multiomics data. To address this gap and propel the field forward, we are proposing to develop computational methodologies that will (1) characterize the state of a single cell in a multiomic setting in an unsupervised manner, (2) characterize the regulatory landscape of single-cells by identifying transcription factor binding activity and (3) identify multiple structural variations at single-cells. First, we will develop interpretable topic-modeling based methods for characterizing single cells based on multiomic readout. Building on our past success with topic models to accurately cluster and characterize single-cell populations, we will develop novel topic modeling approaches for multiomics assays in order to achieve a much deeper profiling of the state of a cell, which will lead to potential insights into the links between multiple modalities measured by a multiomic assay, such as transcriptomic and epigenomic state. Second, we propose to develop a single-cell transcription factor footprinting (TF) methods. Computational detection of TF footprints can identify the landscape of active transcription factors that determine important drivers of cell state and identity. We will develop the methodology to identify active transcription factors at an unprecedented single-cell resolution and to investigate links between the methylome and TF binding. Lastly, we will develop methods to identify different types of structural variations at single cell resolution by leveraging novel multiomic assays developed by my collaborators. This methodology will enhance our ability to study the heterogeneity of structural variations in different cell populations. Overall, the proposed suite of computational methodologies will allow a broad audience of researchers who generate and analyze multiomic data to annotate the multi- modally measured cell states in heterogeneous cell populations in a deep and unprecedented manner at single-cell resolution. Furthermore, our interpretable methods will yield testable hypotheses to better understand the cell state.