Fig. 1

DCATS improves composition analysis through accounting for uncertainty in classification of cell types in differential abundance analysis. A Illustration of the DCATS workflow. Matrices with light blue are input matrices, matrices with light orange are output from DCATS. First step is bias correction using a similarity matrix. This step is optional. The design matrix with multiple covariates is also required. DCATS supports both categorical and continuous covariates. Then, DCATS detects differential abundance using a beta-binomial generalized linear model (GLM) model, which returns the estimated coefficients and p-values. B These box plots illustrate the effect of cell type misclassificaiton in a theoretical simulation with Dirichlet-Multinomial sampling. The similarity matrix is designed to introduce misclassification errors. The proportions of cell types B and C are changed between conditions 1 and 2. C Area under the precision-recall curve (AUC, same below) values when varying p value in detecting the cell type with differential abundance