Accurate highly variable gene selection using RECODE in scRNA-seq data analysis

Accurate selection of highly variable genes (HVGs) is essential in single-cell RNA sequencing (scRNA-seq) data analysis, as it enables the identification of functionally important genes and the characterization of cell types and states. However, HVG selection is often confounded by technical noise inherent in the scRNA-seq measurement process, leading to the misidentification of biologically relevant genes. In this study, we propose a novel HVG selection method based on the RECODE-denoised variance. RECODE is a recently developed noise reduction algorithm that models technical noise in scRNA-seq data as a random sampling process and removes it through a theoretically grounded denoising procedure. Benchmarking on 10x Genomics PBMC data revealed that our method outperformed conventional approaches in capturing known marker genes. Moreover, it demonstrated robustness to changes in normalization parameters and downsampling of cell numbers. Importantly, the same denoised data used for HVG selection can be directly reused in downstream analysis, ensuring consistency throughout the scRNA-seq data analysis pipeline.