De novo gene signature identification from single‐cell RNA‐seq with hierarchical Poisson factorization

• Published in: Molecular systems biology Vol. 15; no. 2; pp. e8557 - n/a
• Main Authors: Levitin, Hanna Mendes, Yuan, Jinzhou, Cheng, Yim Ling, Ruiz, Francisco JR, Bush, Erin C, Bruce, Jeffrey N, Canoll, Peter, Iavarone, Antonio, Lasorella, Anna, Blei, David M, Sims, Peter A
• Format: Journal Article
• Language: English
• Published: London Nature Publishing Group UK 01.02.2019; EMBO Press; John Wiley and Sons Inc; Springer Nature
• Subjects: Artificial intelligence; Bayes Theorem; Bayesian analysis; Budgets; Cell cycle; Datasets; dimensionality reduction; EMBO09; EMBO17; Factorization; Gene expression; Gene Expression Regulation, Neoplastic - genetics; Gene sequencing; gene signature discovery; Genomes; Genomics; Glioblastoma; Glioma; Glioma - genetics; Glioma - pathology; Glioma cells; High-Throughput Nucleotide Sequencing - methods; Humans; Method; Methods; Physiology; Poisson Distribution; Principal components analysis; Ribonucleic acid; RNA; Single-Cell Analysis; single‐cell RNA‐seq; Sparsity; Statistics; Subpopulations; Survival; Temporal variations; Transcriptome - genetics; single‐cell RNA‐seq; glioma; dimensionality reduction; gene signature discovery
• ISSN: 1744-4292; 1744-4292
• DOI: 10.15252/msb.20188557

Common approaches to gene signature discovery in single‐cell RNA‐sequencing (scRNA‐seq) depend upon predefined structures like clusters or pseudo‐temporal order, require prior normalization, or do not account for the sparsity of single‐cell data. We present single‐cell hierarchical Poisson factorization (scHPF), a Bayesian factorization method that adapts hierarchical Poisson factorization (Gopalan et al , 2015 , Proceedings of the 31st Conference on Uncertainty in Artificial Intelligence , 326) for de novo discovery of both continuous and discrete expression patterns from scRNA‐seq. scHPF does not require prior normalization and captures statistical properties of single‐cell data better than other methods in benchmark datasets. Applied to scRNA‐seq of the core and margin of a high‐grade glioma, scHPF uncovers marked differences in the abundance of glioma subpopulations across tumor regions and regionally associated expression biases within glioma subpopulations. scHFP revealed an expression signature that was spatially biased toward the glioma‐infiltrated margins and associated with inferior survival in glioblastoma. Synopsis Single‐cell Hierarchical Poisson Factorization (scHPF) is a Bayesian factorization method for de novo discovery of both continuously varying and subpopulation‐specific expression patterns in single‐cell RNA‐sequencing data. scHPF takes genome‐wide molecular counts as input, avoids prior normalization, captures the statistical structure of scRNA‐seq data better than alternative methods, and has fast, memory‐efficient inference on sparse scRNA‐seq data. Applied to scRNA‐seq of a spatially sampled high‐grade glioma, scHPF reveals regional differences in lineage resemblance within glioma subpopulations. One regionally biased gene signature enriched in astrocyte‐like glioma cells is associated with poor survival in glioblastoma. Graphical Abstract Single‐cell Hierarchical Poisson Factorization (scHPF) is a Bayesian factorization method for de novo discovery of both continuously varying and subpopulation‐specific expression patterns in single‐cell RNA‐sequencing data.