Resolution of the curse of dimensionality in single-cell RNA sequencing data analysis

Abstract

Single-cell RNA sequencing (scRNA-seq) can determine gene expression in numerous individual cells simultaneously, promoting progress in the biomedical sciences. However, scRNA-seq data are high-dimensional with substantial technical noise, including dropouts. During analysis of scRNA-seq data, such noise engenders a statistical problem known as the curse of dimensionality (COD). Based on high-dimensional statistics, we herein formulate a noise reduction method, RECODE (resolution of the curse of dimensionality), for high-dimensional data with random sampling noise. We show that RECODE consistently resolves COD in relevant scRNA-seq data with unique molecular identifiers. RECODE does not involve dimension reduction and recovers expression values for all genes, including lowly expressed genes, realizing precise delineation of cell fate transitions and identification of rare cells with all gene information. Compared with representative imputation methods, RECODE employs different principles and exhibits superior overall performance in cell-clustering, expression value recovery, and single-cell–level analysis. The RECODE algorithm is parameter-free, data-driven, deterministic, and high-speed, and its applicability can be predicted based on the variance normalization performance. We propose RECODE as a powerful strategy for preprocessing noisy high-dimensional data.

Adjustments to the reference dataset design improve cell type label transfer

“…When determining distances between different cell states in high-dimensional space, we often face the challenge of the curse of dimensionality ( Imoto et al, 2022 ). The higher the number of dimensions, the more severe the issue, as noise disproportionately adds up to undermine the true (biological) signal when considering multiple dimensions.…”

Induction of fetal meiotic oocytes from embryonic stem cells in cynomolgus monkeys

“…After putatively excluding doublets/multiplets using the Scrublet python package (v. 0.2.1; Wolock et al , 2019), interspecies doublets composed of mouse somatic cells and cyESC‐derived cells were removed if common cell barcodes were detected in both filtered count matrices mapped to GRCm38.p6 and MacFas5.0. Then, the filtered count matrix was applied to noise reduction by RECODE (resolution of the curse of dimensionality; Imoto et al , 2022) and normalized by 100 k.…”

“…Note that “unclassified” cells in cy fetal ovary germ cells (Mizuta et al , 2022) were not included. Their raw count matrices were again processed by RECODE (Imoto et al , 2022), normalized by 100 k, and then log‐transformed [log (size‐scaled (ss) UMI + 1)]. Batch corrections, identification of highly variable genes (HVGs), and cell clustering were performed as described previously (Satija et al , 2015; Mizuta et al , 2022), and the germ cells were classified into 12 clusters.…”

V-Mapper: topological data analysis for high-dimensional data with velocity

“…Therefore, we used these cell types in the later stage (n = 1, 876 cells and d = 27, 998 genes), pre-endocrine and mature endocrines (α-, β-, δ-, and ε-cells), which were previously annotated in the literature [1,3]. We analyzed the scRNA-seq data as follows (see Appendix B for details): As a pre-analysis of V-Mapper, we used RECODE (resolution of the curse of dimensionality) [12] and RNA velocity [3,14] (Fig. 5).…”

“…The original scRNA-seq data of the pancreas endocrine cell was obtained from the datasets.pancreas function in the scvelo package [3]. We created denoised and velocity data by applying RECODE [12] and RNA-velocity with the stochastic mode parameter [3] to the original data. Next, we extracted data regarding the cells of pre-endocrine and α-, β-, δ-, and ε-cell types from the denoised and velocity data.…”