Single-cell Transcriptome Study as Big Data

Abstract

The rapid growth of single-cell RNA-seq studies (scRNA-seq) demands efficient data storage, processing, and analysis. Big-data technology provides a framework that facilitates the comprehensive discovery of biological signals from inter-institutional scRNA-seq datasets. The strategies to solve the stochastic and heterogeneous single-cell transcriptome signal are discussed in this article. After extensively reviewing the available big-data applications of next-generation sequencing (NGS)-based studies, we propose a workflow that accounts for the unique characteristics of scRNA-seq data and primary objectives of single-cell studies.

Keywords: Big data; RNA-seq; Signal normalization; Single cell; Transcriptional heterogeneity.

Figure 1

Number of papers/datasets addressing single-cell data and big data Searches were performed on January 04, 2016 on http://www.ncbi.nlm.nih.gov/gds for datasets and http://www.ncbi.nlm.nih.gov/pubmed for papers. Data were obtained according to the search criteria as follows filtered by year: (1) for scRNA-seq datasets on GEO: “single cell”[All Fields] AND “Expression profiling by high throughput sequencing”[Filter]; (2) for scRNA-seq papers on PubMed: “single cell”[All Fields] AND (“rna-seq”[All Fields] OR “rna sequencing”[All Fields] OR (“sequencing”[All Fields] AND “transcriptome”[All Fields])); and (3) for big-data papers on PubMed: “big data”[All Fields] OR “hadoop”[All Fields].

Figure 2

MYH2 gene is the marker of mature myotubes The increased bulk expression of MYH2 is primarily driven by the growing proportion of “on-” component cells (upper cluster) over time (0, 24, 48, and 72 h after myoblast differentiation is induced). Figures were derived from the dataset in Trapnell et al . A. The growth of MYH2 expression in bulk cell replicate samples (n = 3 over time). B. Beeswarm plots of the growing bimodal proportion of MYH2 from scRNA-seq over time. C–F. RNA-FISH signals at 0, 24, 48, and 72 h, respectively. MYH2 and nucleus are shown in red and blue (DAPI staining), respectively. Scare bar: 25 nm. G. MYH2 RNA molecule counts per cell over time, based on RNA-FISH analyses. RNA-FISH, RNA-fluorescence in situ hybridization.

Figure 3

Workflow of inter-institutional scRNA-seq data integration Inter-institutional single-cell RNA-seq datasets are aligned against their genomes at the Hadoop layer. Read counts are resolved into gene “on” or “off” status at the normalization layer. Differential expression, co-expression, and other applications are developed based on gene “on” or “off” status instead of gene expression. Biology in the resulting gene list is verified by GSEA, GO-term enrichment analysis, DAVID functional analysis or other tools. GSEA, gene set enrichment analysis; GO, gene ontology; DAVID, database for annotation, visualization and integrated discovery.