Characterizing the impacts of dataset imbalance on single-cell data integration

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) integration methods are crucial for combining datasets.
• Existing methods often fail to account for cell type imbalances across samples.
• Such imbalances can skew downstream analytical results and biological interpretations.

Purpose of the Study:

• To investigate the impact of cell type and cell number imbalances on scRNA-seq data integration.
• To assess the robustness of integration techniques under varying degrees of sample imbalance.
• To develop strategies for characterizing and mitigating the effects of imbalance in scRNA-seq integration.

Main Methods:

• Developed the Iniquitate pipeline to assess integration robustness under imbalance.
• Benchmarked five state-of-the-art scRNA-seq integration techniques across 2,600 experiments.
• Introduced new metrics: aggregate cell type support and minimum cell type center distance.

Main Results:

• Sample imbalance significantly impacts unsupervised clustering, cell type classification, and differential expression analysis.
• Imbalance affects marker gene annotation, query-to-reference mapping, and trajectory inference.
• Quantified the influence of imbalance using novel metrics.

Conclusions:

• Cell type imbalance is a critical factor affecting scRNA-seq data integration outcomes.
• Existing integration methods are sensitive to sample imbalance, impacting biological interpretation.
• Introduced balanced clustering metrics and guidelines to improve integration accuracy and reliability.