Figure 3

Model parameters. (a) Stricter thresholds for target gene calling result in better predictions. (b) Combination of independent histone modification data sets can improve target predictions. Predictions for 203 TFs are evaluated by area under the receiver operator characteristic (ROC) curve (AUC) for data from Pokholok et al. [23] and Kurdistani et al. [22]. (c, d) Using histone modifications within a 500-bp (c) and 1, 000-bp (d) window upstream and downstream of ATG sites of target genes achieves a similar performance to using only upstream signals, and better performance to using only downstream signals. There was no significant performance difference between using 500-bp and 1, 000-bp window sizes. (e) Using histone modifications in intergenic regions has better predictive power than using those in ORF coding regions.