Fig. 4

methylCC: A latent variable model with region-specific and platform-dependent random effects improves estimates of cell composition. Cell composition estimates from N= 10 whole blood samples (K = 6 cell types) measured on the Illumina 450k microarray platform (x-axis) and the RRBS platform (y-axis). Two methods were used to estimate the cell composition: (1) the model proposed by Houseman et al. [3], as implemented by Jaffe et al. [5], that was developed for samples measured on the Illumina 450k microarray platform (left), and (2) our proposed method that is independent of platform technology (right). Root mean squared error (RMSE) is the difference between the cell composition estimates for whole blood DNAm samples measured on the Illumina 450k array platform and on the RRBS platform (averaged across cell types). Within each plot, the RMSE is shown for the reference-based method from Houseman et al. [3] (left) and our proposed model (methylCC) (right)