Presenter: Rahul Satija, Ph.D., New York Genome Center
Abstract: Single-cell RNA-seq has proven to be a transformative tool for the discovery of cell types and states, but the transcriptome reflects only one aspect of cellular identity. The next frontier for single-cell genomics is ‘multi-modal’ analysis, where multiple types of measurements are simultaneously collected from individual cells, often in parallel with gene expression. For example, the CITE-seq technology allows for RNA and protein levels to be simultaneously measured in single cells.
In this talk, I will introduce new statistical methods designed for ‘multi-view’ learning based on multi-modal data. We introduce an unsupervised strategy to quantify the information content related to each modality in each cell, and construct a weighted kernel that integrates information from each source. We apply this approach to a CITE-seq dataset of 250,000 human blood cells with 220 surface proteins, and construct an integrated atlas of the human immune system, and demonstrated. This atlas can be used to map and interpret additional datasets of immune perturbations, for example, to identify specific immune cell types that are affected during response to vaccination, sepsis, or COVID-19.