Abstract
Researchers stand to gain insight into complex biological systems by assembling multiple single-cell RNA-sequencing (scRNA-seq) studies to reveal a panoramic view of overarching biological structure. Unfortunately, many existing scRNA-seq analyses are limited by sensitivity to study-specific noise patterns, by lack of scalability to large datasets, or by integrative transformations that obscure biological relevance. We therefore introduce a novel algorithmic framework that analyzes groups of cells in coexpression space across multiple resolutions, rather than individual cells in gene expression space, to enable multi-study analysis with enhanced biological interpretation. We show that our approach reveals the biological structure spanning multiple, large-scale studies even in the presence of batch effects while facilitating biological interpretation via network and latent factor analysis. Our coexpression-based analysis enables an unprecedented view into two complex and dynamic processes—neuronal development and hematopoiesis—by leveraging a total of seven studies containing 1,460,527 cells from laboratories spanning three continents, yielding systems-level insight unattainable by any individual experiment. Our work demonstrates a path toward probing highly complex biological systems from emerging consortium-scale single-cell transcriptomics.
