Identification of cell types in a mouse brain single-cell atlas using low sampling coverage

Area of Science:

• Single-cell transcriptomics
• Computational biology
• Developmental biology

Background:

• High-throughput single-cell RNA sequencing (scRNA-seq) enables large-scale surveys of cellular diversity.
• Efficiently generating comprehensive cell atlases requires balancing cell sampling numbers with cost-effectiveness.

Purpose of the Study:

• To investigate the relationship between cell numbers sampled and the identification of transcriptional heterogeneity.
• To develop a computational framework for determining optimal cell sampling for scRNA-seq experiments.
• To assess the cost-effectiveness of cell atlas generation strategies.

Main Methods:

• Analysis of a 1.3 million single-cell mouse brain dataset and validation with published data.
• Development of a computational framework to infer cluster discovery saturation points via downsampling.
• Introduction of a "complexity index" to quantify cellular heterogeneity.

Main Results:

• Biologically meaningful cell type distinctions can be recapitulated with significantly fewer cells than originally sampled.
• A computational framework using cluster preservation in downsampled datasets helps infer saturation points.
• While common cell types are robustly identified with fewer cells, extremely rare populations (<1%) require higher cell numbers.

Conclusions:

• Most biologically interpretable cell types in large scRNA-seq datasets can be identified by analyzing approximately 50,000 cells.
• Cell atlas studies may benefit from broader sampling (more individuals/time points) at lower cellular coverage, with targeted enrichment for rare populations.
• This approach optimizes cost and time efficiency, particularly when rare cell populations are not the primary focus.