A computational method to aid in the design and analysis of single cell RNA-seq experiments

Abstract

The advent of single-cell RNA sequencing (scRNA-seq) has given researchers the ability to study transcriptomic activity within individual cells, rather than across hundreds or thousands of cells as with bulk RNA-seq techniques. The greater precision afforded by scRNA-seq identifies mutations and gene expression landscapes private to individual cells or subpopulations, enabling us to determine novel cell types and understand biological systems at greater resolution. Usually biological insights are obtained through the use of unsupervised learning methods on high dimensional single-cell datasets. These methods have to take into account the technical noise structure and distributional properties of scRNA-seq datasets for optimal results. Because the optimal set of analysis methods is different between datasets and there is a wide selection of methods available, it can be both daunting and challenging to design an effective scRNA-seq experiment. In this study, we propose an empirical approach to design a better scRNAseq experiment and answer unresolved biological questions. The tool helps to determine the number of single cells to be profiled and the optimal computational pipeline based on the characteristics of the tissue system under study. Using simulated datasets, we demonstrate that the number of single cells required and the appropriate analysis strategy depend on the characteristics of the cell types under investigation1.