Fig. 2

Assessing model fit. Predicted proportions from the model are plotted against the DNA methylation-derived cell proportions for each sample in the training data (a) or that obtained from CBC/Diffs (b). For (a), linear best-fit line to the data is plotted (blue line) with 95% point-wise confidence interval for fit (grey band) and compared with perfect agreement (red dashed line). For (b), predicted monocyte proportions are compared directly to the CBC/diffs. The predicted granulocyte proportions are compared to the sum of neutrophil, eosinophil and basophil proportions from the CBC/diffs, while the sum of the predicted B, CD4+ T, CD8+ T and NK cell proportions is compared to the total lymphocyte proportions from the CBC/diffs. For each cell type, Pearson’s product–moment correlation (Pearson’s r) and the root mean squared error (RMSE) are reported