We also evaluate the effectiveness of our scoring scheme in finding binding sites for five regulatory proteins when they are embedded in simulated data. Our goals are twofold. First, since our underlying scoring measure is based on counting matches between nucleotides, it is important to see how well it performs in compositionally biased backgrounds. In the E. coli dataset, even a simple scoring scheme that assigns a score of 1 to matches and 0 to mismatches performs competitively (data not shown). However, since other genomes can have considerably more biased nucleotide compositions, our scoring scheme rewards matches between more rare nucleotides, and we test here how it performs in different scenarios. Second, while it is essential to test the performance of motif finding algorithms on genomic data (as above), it is possible that there are other conserved motifs in the data, besides those with which we are evaluating performance, and these conserved motifs lead to lower nPC and sSn measurements. Simulated data is not expected to have other conserved motifs, and thus provides a cleaner, though perhaps optimistic, means for testing motif finding approaches.