Parameter Landscape Analysis for Improving the Performance of Common Motif Detection Algorithms

It is known that genes with similar expression profile are likely to be regulated by a common transcription factor and finding the common cis-element to which the protein binds from the upstream regions of these genes is very important. Although several famous programs for motif extraction exist nowadays, end-users often hesitate about the reliability of the output results. One reason is that the user does not have any a priori knowledge about the motif, i.e., either it is a single motif, multiple similar motifs, or multiple non-similar motifs; even some of the sequence may not contain any motifs at all. Although the performance of some of these famous programs, such as MEME [1], Gibbs DNA [3] and Consensus [5], has been studied [2, 4, 6], they were usually done in their default mode, which cannot be suited for all the circumstances. In our research, we systematically explored various possibilities of parameter-setting, which is the most complicated part of these programs in realistic situations. In this way, we tried to solve the problem of elucidating strongly corrupted motifs, such as 10/2, 12/3, 15/4 (motif length/mismatches), which have been reported before as being very difficult with the above algorithms [4].