paperKB
coga / coga-kb
Help
Sign in

Chunk #21 — Results — scCODA accounts for the negative correlation structure for compositional changes and shows fewer false positives

Source
scCODA is a Bayesian model for compositional single-cell data analysis.
Embedded
yes

Text

Our final analysis considered a longitudinal scRNA-seq dataset from the small intestinal epithelium in mice, studying the effects of Salmonella and Heligmosomoides polygyrus infection on cell-type composition6. In contrast to the original Poisson regression data analysis6, scCODA found only a single statistically credible increase in Enterocytes in Salmonella infected mice for an FDR level of 0.2 (Supplementary Fig. 10 and Supplementary Data 5). In addition, the Poisson model identified Tuft cells to be significantly affected after three and ten days of infection with H. polygyrus, while Enterocytes, Goblet, and early transit-amplifying cells were found to change significantly only after ten days of infection (Supplementary Fig. 10). All these changes could not be confirmed by scCODA at an FDR level of 0.2. For comparison, ANCOM did not find any significant changes for all three conditions, confirming its lack of power for datasets with few samples.