Statistical Method mcRigor Enhances the Rigor of Metacell Partitioning in Single-Cell Data Analysis

Single-cell sequencing data suffers from sparsity and noise, which researchers address by aggregating similar cells into 'metacells.' A new statistical method called mcRigor enhances the reliability of this process by detecting dubious or internally inconsistent metacells within a given partition. The method uses a feature-correlation statistic, mcDiv, to quantify heterogeneity and a double permutation procedure to identify metacells that deviate from a null distribution. mcRigor also provides a scoring system to select the optimal metacell creation method and hyperparameters for a given dataset, improving downstream analyses like gene co-expression and regulatory inference.