Figure 2

Accurate models of higher-order chromatin state built from locus-level features.(A) Model predictions (Predicted eig) are compared to observed values (Empirical eig). Various metrics are used to measure the accuracy of regression modeling – the Pearson correlation coefficient (PCC) and root mean-squared error (RMSE) – and to evaluate classification accuracy (for A e i g≥0 and B e i g<0) – accuracy (percentage of true positives, Acc.) and area under the receiver operating characteristic curve (AUROC). (B) Variable importance (shown for the ten most informative features per model) is calculated as the decrease in the accuracy of predictions for the permuted variable relative to observed (in units of percentage increase in MSE), averaged over the forest (see Materials and methods). Acc., accuracy; AUROC, area under the receiver operating characteristic curve; eig, eigenvector; MSE, mean-squared error; PCC, Pearson correlation coefficient; RMSE, root mean-squared error.