Hierarchical Canonical Correlation Screen for Identification and Visualization of Phenotype-Driven Cell Types in Multiple Sample Single-Cell RNA-Sequencing Experiments

• Activity Number: 55 - Statistical methods for data from single cell technologies
• Type: Contributed
• Date/Time: Sunday, August 8, 2021 : 3:30 PM to 5:20 PM
• Sponsor: Section on Statistics in Genomics and Genetics
• Author(s): Boyang Zhang* and Hongkai Ji
• Companies: Johns Hopkins Bloomberg School of Public Health and Johns Hopkins Bloomberg School of Public Health
• Keywords: single cell genomics; visualization; canonical correlation; hierarchical tree; big data; COVID-19

Abstract

With the rapid development of single-cell RNA-sequencing (scRNA-seq) technology, more and more multi-sample scRNA-seq datasets become increasingly common. An emerging challenge is to conveniently explore and visualize the complex information about association between sample phenotype and gene expression. Here we propose a new framework based on canonical correlation that systematically examine transcriptomic features across different hierarchies and quickly pinpoint key cell types that are pertinent to phenotypes of interest. We provide a flexible end-to-end pipeline to build hierarchical tree (e.g. using existing known biology or data-driven approach), summarize features (e.g. cell type proportion or transcriptomic signature) at each hierarchical level in multiple ways (e.g. via concatenating or aggregating) and highlight phenotype-related cell types. We demonstrate our method by analyzing an integrative single-cell COVID-19 PBMC cell atlas and a scRNA-seq dataset from neoadjuvant immunotherapy clinical trial for non-small cell lung cancer.