Fig. 1

Schematic representation of the experiment. Cell proportions were estimated from DNA methylation profiles for 172 samples (1; COPD patients). These DNA methylation-derived cell proportions were used as a ‘silver’ standard in the absence of the ground truth. This ‘silver standard’ dataset was used to train a multi-response Gaussian model using the gene expression data (2). Out-of-sample performance was evaluated using a repeated (20x) 10-fold cross-validation (3) and in two independent sets of samples, in different clinical indications: granulocyte, monocyte, and lymphocyte performance was evaluated in a large set of samples (4; 197 samples, heart failure), while monocyte, B, and T cell prediction performance was evaluated in a smaller set (4; 28 samples, asthma)