How many markers are needed to robustly determine a cell's type?
Stephan Fischer1, Jesse Gillis1,2
1Cold Spring Harbor Laboratory, Stanley Institute for Cognitive Genomics, Cold Spring Harbor, NY 11724, USA.
Iscience
|November 12, 2021
Summary
Researchers developed a framework to identify reliable marker genes for cell types using multiple datasets. This helps in accurately annotating and understanding cells, especially rare ones.
Area of Science:
- Neuroscience
- Genomics
- Computational Biology
Background:
- Single-cell atlases have significantly advanced cell type understanding.
- Marker genes are crucial for experimental validation and computational analyses like cell annotation and deconvolution.
- A standardized method for quantifying marker gene replicability is currently absent.
Purpose of the Study:
- To develop a framework for quantifying marker gene replicability.
- To identify robust and replicable marker genes for 85 neuronal cell types from the Brain Initiative Cell Census Network (BICCN).
Main Methods:
- Systematic investigation of marker replicability using high-quality BICCN data.
- Combining data from 5 datasets to achieve robust differentially expressed (DE) gene identification.
- Meta-analysis to determine optimal numbers of markers for downstream applications.
Main Results:
- Dataset-specific noise necessitates combining multiple datasets for reliable marker identification, especially for rare and lowly expressed genes.
- An estimated 10 to 200 meta-analytic markers offer optimal performance for downstream tasks.
- Replicable marker lists for 85 BICCN neuronal cell types are now available.
Conclusions:
- Replicable marker lists provide interpretable and generalizable cell type information.
- These lists facilitate downstream applications such as cell type annotation, gene panel selection, and bulk data deconvolution.
- The developed framework addresses the need for robust marker gene selection in single-cell genomics.
