Implications of clustering in cell type annotation with single cell RNAseq data

Abstract

Single cell RNA-sequencing (scRNAseq) provides high-resolution data necessary to investigate rare cell populations contributing to treatment resistance commonly observed in many forms of cancer. Generally, the first step in this investigation is understanding the cellular makeup of a tissue sample by annotating cells based on their RNA expression profile. In this study, we compare cluster-based cell type annotation methods with approaches that annotate cells individually without employing clustering. Cell type frequencies are identified across 22 ovarian cancer tumor samples sequenced using 10x Genomics single cell RNA sequencing. These approaches are compared to identify biases, enable the identification of rare cell populations, and ultimately allow the correlation of cellular profiles to clinical response with the long-term goal of using these data to tailor treatment options and improve patient outcomes.