A k-mer Support Vector Machine (SVM) can accurately predict H3K27me3 domains from the primary sequence in vertebrates. (A) Examples of H3K27me3 domain prediction by a k-mer SVM using k = 8 (lower track; black is a positive score, gray a negative score). H3K27me3 enrichment in the blastula and gastrula is visualized in the top two tracks. (B) Interspecies performance of k-mer SVMs (k = 8) trained on different vertebrates. Performance is visualized in a receiver operator curve (ROC) and the ROC area under curve (ROC AUC) is shown as a performance measure. The y-axis shows the sensitivity; the x-axis, 1 − specificity. The columns indicate the species used for training; the rows indicate the species used for evaluation. For instance, the lower left ROC curve indicates the performance of human sequence classification of an SVM trained on frog.
