Fig. 3

Evaluation of reference marker selection in each deconvolution method. A Upset plots illustrate the intersections of selected markers from various methods. cfNOMe used the same set of reference markers as MethAtlas. B Markers’ specificity measured by absolute methylation level difference between the target cell type and all the others. Data are represented as mean ± standard deviations (SD). C Markers’ specificity measured by statistical difference between the target cell type and all the others. Statistical difference was calculated as -log₁₀(P-value) using Student t-test. Data are represented as mean ± SD. D The variability of the selected markers across healthy individuals was measured by interquartile range (IQR). IQR data were sourced from the ImmuMethy database, focusing on four immune cell types. E The variability of the selected markers across diverse healthy states was measured by IQR. Data for IQR were calculated based on data extracted from the ImmuMethy database. F Normalized performance score across 25 combinations of marker selection and cell type proportion estimation modules. The median normalized performance score, calculated from 100 in silico cfDNA samples, was represented by color and labeled accordingly. Rows represent cell type proportion estimation methods, and columns represent marker selection methods. P-values between different methods in B–E were determined using Wilcoxon rank-sum test. **P < 0.01, ***P < 0.001