Response to 'Comments on "MMFPh: A Maximal Motif Finder for Phosphoproteomics Datasets"'

Recently, two new approaches to find overrepresented motifs in phosphoproteomics datasets have been introduced: our MMFPh (Wang et al., 2012) and He et al.’s Motif-All (BMC Bioinformatics 2011). Both methods espouse the importance of completeness— finding all motifs supported by the data—in contrast to previous approaches that may miss some motifs due to algorithmic choices. As we discuss in the Introduction of our article, however, while both methods seek to identify all significant motifs, they employ different significance assessments. In some cases, the difference does not matter much if at all. However, as we show in the Results, in some cases the difference leads to Motif-All finding many more motifs than MMFPh, and many more than those that are biologically supported, including known false positives planted in synthetic datasets. They also lead to Motif-All occasionally missing a motif found by MMFPh, though not for the datasets and parameter settings employed in the presented examples. Since MMFPh and Motif-All employ different notions of significance, it is not surprising that empirically they do not find exactly the same sets of motifs. He et al. (submitted for publication) elaborate on this finding by providing a theoretical characterization with respect to their notion of significance, which is a global assessment of an entire peptide. In contrast, MMFPh employs a local assessment of individual amino acid/position pairs during construction of a motif [our Equation (1)], as introduced by the popular Motif-X approach to phosphorylation motif discovery