Fig. 1

Epigenetic signature of Gleason score in The Cancer Genome Atlas. a Variable selection using elastic net. Each curve in the figure corresponds to a single CpG site. It shows the path of its coefficient (y-axis) against the log(lambda) or tuning parameter (x-axis). The vertical dashed line represents the optimal log(lambda) for classifying high (8−10) versus low (≤6) Gleason score tumors, which was identified using cross-validation. Based on an optimal log(lambda) of −1.7061, 52 CpG sites were selected. These 52 CpGs and their elastic net coefficients were then used to calculate the epigenetic signature as described in the “Methods” section. b Heatmap of the 52 CpG sites that were selected using elastic net. The rows of the heatmap are the CpG sites, and the columns are the tumor samples. The samples were grouped by Gleason score. Methylation β values (range 0−1) were used, and the highest methylation levels are shown in red. The number of patients with Gleason ≤6, 7(3 + 4), 7(4 + 3), and 8−10 tumors is 65, 102, 78, and 88, respectively. The rows were clustered based on Euclidean distance. c Box plots of the epigenetic signature (y-axis) for patients with different Gleason scores. The signature is presented as a proportion. Higher Gleason scores were associated with higher levels of the signature. The same patients as in Fig. 1b were used