Fig. 6

Key methylation sites describe inter-patient expression variation. a Development of methylation-based models of gene expression variation. b Example models of gene expression variations. Left: Methylation versus expression of the sites consisting the best model of the TNFAIP3 gene. Right: Predicted versus observed expression levels of the TNFAIP3 or the SMO genes across the tumors. SMO model was based on the four sites shown in Fig. S8a. c Fractions of gene-expression models which incorporate methylation data from, positive, negative, or both positive and negative units. d Assessing the prediction power of gene-expression models. Left: Developing of a prediction model for each of the 24 GBM tumors, based on the other 23 tumors. Middle: Example distribution of predicted versus observed expression levels of cancer genes. The full list of distributions is given in Additional file 5. Right: Distributions of predicted versus observed expression levels of driver genes across the 24 tumors. Log 2 of the differences between predicted and observed gene expression levels for driver genes with developed models are shown. Box plots describe the distributions of prediction errors in 24 independent tests