Artificial variables help to avoid over-clustering in single-cell RNA sequencing.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) is crucial for identifying cell types.
• Unsupervised clustering and differential expression analysis are standard steps in scRNA-seq pipelines.
• Over-clustering can lead to misleading results in downstream analyses.

Purpose of the Study:

• To introduce "recall" (calibrated clustering with artificial variables), a novel method to prevent over-clustering in scRNA-seq data.
• To address the issue of "double dipping" in differential expression analysis following clustering.
• To provide a robust and efficient tool for analyzing large-scale scRNA-seq studies.

Main Methods:

• Developed "recall," a method that controls for the impact of data reuse in differential expression analysis.
• Applied recall to various clustering algorithms.
• Validated the method using both real and simulated scRNA-seq datasets.

Main Results:

• Recall effectively protects against over-clustering, improving the reliability of cell type identification.
• The method demonstrates state-of-the-art clustering performance.
• Recall enables rapid analysis of large-scale scRNA-seq studies, even on standard hardware like a personal laptop.

Conclusions:

• Recall is a valuable tool for enhancing the accuracy and efficiency of scRNA-seq data analysis.
• The method mitigates common pitfalls in scRNA-seq pipelines, particularly over-clustering and double dipping.
• Recall facilitates robust biological discovery from complex single-cell datasets.