InCoB 2020 Keynote and Plenary Speakers
Shyam PrabhakarSenior Group Leader Agency for Science Technology and Research (A*STAR)Laboratory of Systems Biology and Data Analytics Associate Director, Spatial and Single Cell Systems
Shyam Prabhakar obtained a B.Tech in Electronics and Communications Engineering from the Indian Institute of Technology, Madras and a PhD in Applied Physics from Stanford University. He was sole recipient of the 2001 American Physical Society PhD thesis award for Beam Physics. Following postdoctoral fellowships in Mathematics at Stanford University and Genomics at the Lawrence Berkeley National Laboratory, he joined the Genome Institute of Singapore to start his own research group. His lab uses single-cell assays, spatial omics, epigenomics and novel algorithms for basic science and translational studies of disease processes. Major achievements include the first single-cell transcriptomic analysis of colorectal cancer, the first large-scale screen for histone acetylation QTLs, the first histone acetylome- wide association study of a psychiatric disorder and the first general-purpose peak detection algorithm for omics profiles. He has led multiple industry collaborations and served on or chaired multiple grant panels, conference organizing committees, task forces and steering committees related to single cell research, epigenomics and Precision Medicine. He co-founded Human Cell Atlas-Asia and is currently Associate Director, Spatial and Single Cell Systems (GIS) and head of genome analytics for Singapore’s National Precision Medicine Program.
Abstract: Single-cell Data Analytics: Asian Immune Diversity and Cancer Cell States
In recognition of the fact that cellular properties can vary systematically across individuals, the Human Cell Atlas-Asia consortium has initiated a flagship single cell project to generate an Asian Immune Diversity Atlas (AIDA). AIDA will profile transcriptome and epigenome variation in peripheral blood within and across population groups, and thereby characterize the influence of age, sex, genetics and environment on immune cell types in Asia. To execute such a geographically dispersed study without succumbing to lab-specific biases, we have standardized
protocols across the consortium, centralized primary data processing and leveraged a common set of control samples across study sites. To reduce cost and minimize technical variation, we pool samples before single cell encapsulation (mux-seq). Early results suggest that these measures yield data that can seamlessly be integrated across the three initial sites.
Cohort-scale single-cell analysis requires industrial-strength data analysis, i.e. scalable algorithms that work uniformly well on all datasets at constant parameter settings. We have therefore developed next-gen algorithms for clustering cells by major cell type (RCA2) and then performing feature selection (DUBStepR) to sensitively identify subtypes within each major cell type. RCA2 combines the robustness of reference-based (supervised) clustering with the scalability of graph-based methods to detect common and rare cell types in large scRNA-seq datasets, despite profound batch effects and technical variation. DUBStepR exploits gene-gene correlations and a novel measure of cell clumping in expression space to identify the optimal gene set for sensitive and accurate cell clustering. Importantly, DUBStepR also generalizes to data types such as scATAC-seq that resist conventional feature
We used these methods to characterize molecular states in colorectal cancer, based on scRNA-seq analysis of >200,000 cells from tumor and matched normal. Notably, each patient contributed at least one unique epithelial cell cluster, indicating a remarkable diversity of cancer cell states across patients. In contrast, stromal and immune clusters were consistently shared across patients. Intriguingly, the clear distinctions between stromal cell types in normal colorectal tissue were smeared into an unbroken continuum of cell states in tumors. In a similar single cell study of bone marrow samples from chronic myeloid leukemia, we found that prevalence of certain cell states at diagnosis was strongly predictive of response to tyrosine kinase inhibitors.