Extended Data Fig. 5: Validation of a machine learning model to predict heart specific enhancers across vertebrates.

5 (a) Parameter tuning to train the SVM with RBF kernel with a grid-search for parameters c and gamma showing the calculated AUC after 5-fold cross validation. AUC = Area under the ROC curve. (b) ROC curves with computed AUC showing the performance of gkm-SVM with either RBF(rbf, orange) or weighted RBF(wrbf, blue) kernel on test data. The SVM was trained with the c & gamma parameters chosen in (a). (c) ROC curves with computed AUC showing human-chicken interspecies prediction accuracy for different conservation classes of mouse promoters projected to chicken. (d) Estimation of sequence alignability as a function of SVM predicted tissue-specificity (as prediction score) for ATAC-Seq peaks from chicken embryonic heart. (e) Top 10 mouse (left) and chicken (right) patterns discovered by TF-MoDisco showing seqlet as CWM, trimmed and converted PWMs and their annotated JASPAR motif match.