Number of predicted motifs and number of matches in ScerTF at 15% FDR, for RED2 (mutual-information scoring function; same as Figure 8), FIRE and MatrixREDUCE. FIRE was run with default parameters (optimized for yeast), with k = 5, 10, 20, 40 and 80 clusters, and the number of clusters that yields the highest number of motifs was selected a posteriori for each dataset. MatrixREDUCE was run with default parameters, with seeds of length 7, 8 and 9. (Left) Average results of the three methods on the 24 yeast datasets. (Center) RED2 and FIRE. Number of predicted motifs that match a known motif at 15% FDR in the ScerTF database for the 24 yeast datasets. The y-axis corresponds to the number achieved by RED2 and the x-axis to the number achieved by FIRE with the best clustering procedure. Superimposed points are indicated by shading. RED2 has more matches than FIRE in 21 datasets and fewer in three datasets, which gives a sign test P value of 0.0003. (Right) RED2 and MatrixREDUCE (same explanations as for the center panel). RED2 has more matches than MatrixREDUCE for 22 datasets and is on par for the remaining two, which gives a sign test P value of 4.77 × 10-7.
Lajoie et al. Genome Biology 2012 13:R109 doi:10.1186/gb-2012-13-11-r109