Fig. 3

Receiver-operating characteristic (ROC) curve using the Bayesian posterior probability (Eq. 8) and the permutation test p-value (Eq. 15) as predictors of eggNOG/COG regulation. The ROC was generated on a synthetic dataset of 10,000 eggNOG/COGs, each with 100 promoter sequences mapping to it. To compute p-values, 100 permuted models were generated. The synthetic dataset contained 100 “regulated” eggNOG/COGs. To simulate real conditions, promoters mapping to “regulated” eggNOG/COGs were assigned sites following the CsoR motif based on a geometric distribution with an expectation of 0.33 sites per promoter