Aspects of single-cell RNA sequencing (scRNA-seq) data, such as its sparsity and high cell–cell variability, have made it hard to apply statistical approaches developed for bulk RNA-seq analyses, such as differential gene expression Kim et al. present Memento, a tool that uses methods-of-moments estimators of the hypergeometric sampling process to estimate single-cell gene abundances, variation and covariation while decomposing biological and technical variability, combined with an efficient bootstrapping scheme to quantify uncertainty and test hypotheses. Having demonstrated the increased power of Memento while maintaining statistical calibration compared with existing methods in simulated differential expression data, the authors showcase analyses using real data, such as differential gene expression and variation, calculations of gene–gene correlation, single-cell CRISPR perturbation screening, and mapping of expression quantitative trait loci (eQTLs). Notably, the computational efficiency of Memento means it can be applied to population-scale scRNA-seq cohorts, and the authors implement a fast pre-computed mode in the CELLxGENE database to compare arbitrary groups of cells. Although single-cell genomics has flourished, the statistical methods to fully utilize these rich but complex data have been lacking — and hence Memento’s breadth of application may be an especially useful step forward.

Original reference: Cell 187, 6393–6410.e16 (2024)

单细胞 RNA 测序 （scRNA-seq） 数据的各个方面，例如其稀疏性和高细胞间变异性，使得难以应用为大量 RNA-seq 分析开发的统计方法，例如差异基因表达 Kim 等人提出了 Memento，这是一种使用超几何采样过程的矩量估计器来估计单细胞基因丰度、变异和协变异的工具，同时分解生物和技术变异， 结合高效的引导方案来量化不确定性并检验假设。在模拟差异表达数据中证明了 Memento 在保持统计校准的同时具有更高的功能，作者展示了使用真实数据的分析，例如差异基因表达和变异、基因-基因相关性的计算、单细胞 CRISPR 扰动筛选和表达数量性状位点 （eQTL） 的定位。值得注意的是，Memento 的计算效率意味着它可以应用于群体规模的 scRNA-seq 队列，并且作者在 CELLxGENE 数据库中实现了一种快速的预计算模式来比较任意细胞组。尽管单细胞基因组学已经蓬勃发展，但一直缺乏充分利用这些丰富而复杂的数据的统计方法——因此 Memento 的广泛应用可能是向前迈出的特别有用的一步。

原始参考：Cell187， 6393–6410.e16 （2024）