Fig. 3

M2A performance in NBL and RMS cohorts. a, b Analysis of the performance of M2A in NBL and RMS cohorts with a H3K4me3 inference and b H3K27ac inference. ENCODE replicate consistencies were calculated as Pearson’s correlation squared (R²) between replicates (two replicates per sample). ENCODE sample KMS-11 was excluded as an apparent outlier R² = 0.016, RMSE = 1.869. Prediction accuracy was measured by R² between the actual measurement and M2A’s prediction. c–g Individual examples of median M2A performers in c NBL cell line H3K27ac inference, d NBL cell line H3K4me3 inference, e RMS H3K27ac inference (pre-transfer), f RMS H3K4me3 inference, and g RMS H3K27ac inference (post-transfer). To indicate the density of data points where individual data points cannot be resolved, a KDE was applied, called from 1 (highest) to 0 (lowest). h The boost to M2A performance (measured by RMSE) due to transfer learning is shown, as applied and tested in RMS samples