PT  - JOURNAL ARTICLE
AU  - Fang Wang
AU  - Tapsi Seth
AU  - Shaoheng Liang
AU  - Nicholas Navin
AU  - Ken Chen
TI  - SCMarker: ab initio marker selection for single cell transcriptome profiling
AID  - 10.1101/356634
DP  - 2018 Jan 01
TA  - bioRxiv
PG  - 356634
4099  - http://biorxiv.org/content/early/2018/07/04/356634.short
4100  - http://biorxiv.org/content/early/2018/07/04/356634.full
AB  - Current single-cell RNA-sequencing (scRNA-seq) data generated by a variety of technologies such as DropSeq and SMART-seq can reveal simultaneously the mRNA transcript levels of thousands of genes in thousands of cells. Cell subpopulations (e.g., cell-types) that have similar transcriptomes can be further delineated in the high dimensional gene expression space. However, genes are not equally informative in delineating cell subpopulations. Therefore, it is often important to select informative genes or subpopulation-informative markers (SIMs) to reduce dimensionality and achieve informative clustering. Here, we present an ab initio method that performs unsupervised marker selection, based on two novel metrics 1) discriminative power of individual gene expressions and 2) mutually coexpressed gene pairs (MCGPs). Consistent improvement in cell-type classification and biologically meaningful marker selection are achieved when applying SCMarker on data generated by scRNA-seq datasets, including UMI data by the 10X Chrimium and TPM data by SMART-seq2, from various tissue types (melanoma, brain, etc.), followed by a variety of clustering algorithms such as k-means, shared nearest neighbor (SNN), etc. The R package of SCMarker is publicly available at https://github.com/KChen-lab/SCMarker.