Performance evaluation of DECA on simulation datasets. (A) The schematic diagram depicts the single-cell ATAC datasets from various tissue: PBMC, thymus, liver and heart. (B) The scatter plot represents correlation within PBMC samples. The x-axis represents the cell-type proportion of ground truth, while the y-axis represents the predicted proportion. (C) The heatmap illustrates the spearman correlation between ground truth and reconstructed chromatin accessibility matrices for PBMC samples. P-values were calculated using a two-tailed Spearman rank correlation test. ^*P < 0.05; ^**P < 0.01; ^***P < 0.001; n.s., not significant. (D) Bar plots of DECA accuracy: Lin's concordance correlation coefficient (CCC) and mean absolute error (MAE), where higher CCC and lower MAE indicate better performance (see Methods). (E) These scatter plots of correlations for pseudo-bulk samples using various deconvolution methods (Bisque, DWLS, CIBERSORTx, TAPE, Cellformer) across multi-tissue datasets. (F) These box plots of CCC and MAE for different methods (Bisque, DWLS, CIBERSORTx, TAPE, Cellformer) across multi-tissue datasets. (G) These bar plots of predicted proportions in pseudo-bulk ATAC datasets with different distributions (yellow for actual, blue for predicted).