Fig. 3
From: MethylNet: an automated and modular deep learning approach for DNA methylation analysis
Results on test set (n = 144) for cell-type deconvolution: a For each cell type, the predicted cellular proportion using MethylNet (x-axis) was plotted against the predicted cellular proportion using estimateCellCounts2, which has been found to be a highly accurate measure of cellular proportions and thus serving as the ground truth for comparison, a regression line was fit to the data for each cell type: B-cell, CD4T, CD8T, Monocytes (Mono), NK cells, and Neutrophils (Neu); b Grouped box plot demonstrating the concordance between the distributions of the MethylNet-estimated proportions of each cell-type and the distributions derived using estimateCellCounts2; c Hierarchical clustering using the correlation distance between two cell types’ SHAP CpG scores across all CpGs. The linkage is found between cell types of similar lineage