A systematic evaluation of highly variable gene selection methods for single-cell RNA-sequencing

Selecting highly variable genes (HVGs) is a critical step in single-cell RNA sequencing data analysis. We benchmark 47 HVG selection methods across 19 datasets, 18 evaluation criteria, and 5,358 settings. Hybrid methods – mixtures of multiple baseline HVG approaches – robustly outperform individual methods. Based on these findings, we develop mixHVG, an improved HVG selection strategy that integrates top-ranked genes from multiple baseline approaches. To facilitate its use, we provide an open-source R package, mixhvg, designed for easy integration into existing analysis pipelines. Our benchmark framework also offers a valuable resource for future method development and evaluation.