Skip to main content

Advertisement

Fig. 2 | Genome Biology

Fig. 2

From: BayesCCE: a Bayesian framework for estimating cell-type composition from DNA methylation without the need for methylation reference

Fig. 2

The performance of existing reference-free methods and BayesCCE under the assumption of six constituting cell types in blood (k=6): granulocytes, monocytes, and four subtypes of lymphocytes (CD4+, CD8+, B cells, and NK cells). For each method, box plots show for each data set the performance across ten sub-sampled data sets (n=300), with the median indicated by a horizontal line. For each of the methods, ReFACTor, NNMF, MeDeCom, and BayesCCE, we considered a single component per cell type (see the “Methods” section). Additionally, we considered the scenario of cell count imputation wherein cell counts were known for 5% of the samples (n=15; BayesCCE imp) and the scenario wherein samples from external data with both methylation levels and cell counts were used in the analysis (n=15; BayesCCE imp ext). Top panel: mean absolute correlation (MAC) across all cell types. Bottom panel: mean absolute error (MAE) across all cell types. For BayesCCE imp and BayesCCE imp ext, the MAC and MAE values were calculated while excluding the samples with assumed known cell counts

Back to article page