Fig. 4
Promoter bias is evident in predicted sequences. a: Random Forest classifier, as well as SVM based Erwin et al. predictions, show high preference to promoters. * = p-value <10−100, ** = p-value <10−250. b: Random Forest can distinguish enhancers (TSS-containing excluded) and promoters (windows containing TSS) with high AUC, using only 4-mers, placing random sequences scores close to enhancers scores. c and d: Values returned by two-step heart (c) and brain (d) 4-mers classifier for different sets of sequences