Case study : Improved motif detection performance of MotifSampler with a conservation-based PSP on yeast datasets.
===========================================================================

APPENDIX-1 - parameter settings:
-------------
RUNNING MOTIFSAMPLER:
For parameter -b we select the precompiled background model of order 1 for S. cerevisiae from our server, motif detection is done on the positive strand (-s=0), we search for maximal one instance (-M) of one motif (-n=1) in 100 runs (-r=100), with a prior bias to find mainly one instance per sequence (-p=0.9_0.25), and the motif width (-w) is extracted from the known literature motifs used in the Gordan study. We use MotifRanking (minimum overlap parameter -x = (-w)/2 and maximal shift -s = -w, other parameters default) to retain the motif with the highest likely score that was found at least 10 times on 100 runs. The retained motif is compared with the known literature motif using MotifComparison (same setting for -x and -s as in MotifRanking) and the algorithm for a given trial is found successful if the KL score is lower than 0.7 (-t).

APPENDIX-2 - RESULTS ON 62 DATASETS:
-------------
The table lists the successful (1) or unsuccessful (0) motif detection by MotifSampler, PRIORITY and 7 other motif finders in 62 datasets. Successful means that the predicted motif is found similar to the literature motif. Unsuccessful means the opposite or the motif finder may not have found any motif.

End of page.