Fig. 5

Replication from our previous study [3] with GenoGAM 2.0. a Replication of figure 3A from our previous study [3]. ROC curve based on a quantile cutoff of 0.15 (15% of the genes are assumed to be true negatives). GenoGAM (orange and blue) has a constantly higher recall with a lower false positive rate. b Replication of Fig. 3b from our previous study [3]. Area under the curve (AUC) for all possible quantile cutoffs from 0 to 1 in steps of 0.01. GenoGAM 1.0 (blue) and GenoGAM 2.0 (orange) are almost identical and are thus largely overlapping. Up to a cutoff of 0.6, GenoGAM (orange and blue) performs consistently better than all competitor methods by around 0.03-0.04 points above the second best method (csaw and DESeq2, pink and green, respectively). The entire range of quantile cutoffs is shown out of completeness, reasonable values are between 0.15 and 0.25. c Replication of supplementary figure S9C from our previous study [3]. Proportion of significant peaks within 30 bp of motif center and 95% bootstrap confidence interval (error bars) for six ENCODE transcription factors (CEBPB, CTCF, USF1, MAX, PAX5, YY1) on chromosome 22 and for the yeast TFIIB dataset