A new statistical method called crumblr analyses changes in cell type composition from single-cell data with more power than existing tools, according to a paper in Nature Communications.
Single-cell technology now lets researchers count how many cells of each type appear in a sample, across large groups of people. The paper introduces crumblr, a statistical method built to analyse these cell-count ratios and find real differences in composition between groups, such as healthy versus diseased tissue.
Crumblr uses precision-weighted linear mixed models that can include random effects for complicated study designs, and it tests for differences at several levels of the cell lineage hierarchy at once, rather than one cell type at a time.
The authors report that crumblr increases statistical power compared with existing methods while keeping the false positive rate under control, but this is shown through simulations and through four published datasets covering aging, tuberculosis infection in T cells, prostate cancer bone metastases, and SARS-CoV-2 infection. The paper does not claim performance beyond these tested conditions.
For researchers and clinicians working with single-cell data, the practical takeaway is that crumblr offers a way to test cell composition changes across a hierarchy of cell types, not just single categories, which may be useful when cell lineage structure matters to the biological question. As with any new statistical tool, independent testing on further datasets would help confirm how broadly its power gains hold.
