Eliminating repetetive motifs found by MEME

Question

[rbronste]

Whenever I run MEME I get several top E-value hits that are pure repeats (often 25bp or longer) and I would like to eliminate these from the search and keep only actually possible TF motifs. Is there a way to do this? Thanks.

Answer 1

[Alex Reynolds]

Unless you're looking for de novo motif models, one option is to use TOMTOM to rank MEME hits by nearness to published TF databases. That should clean things up considerably.

Unless you're looking for really long binding sites (dimers, say) you could also set the -maxw parameter in MEME so that you're looking for sites that are less than 25nt long. Tuning other parameters may be of use.

Another option is to adjust your background, by removing sequences from repeat-masked regions.

[rbronste]

Great suggestions. Do you know of a good source to obtain a mouse (mm10) bed file of repeat-masked regions? UCSC I guess?

[Alex Reynolds]

UCSC would be my first stop. Others might suggest Biomart, maybe.

$ wget -qO- http://hgdownload.cse.ucsc.edu/goldenPath/mm10/database/rmsk.txt.gz | gunzip -c | awk -v OFS="\t" '{ print $6,$7,$8,$11,$2,$10; }' | sort-bed - > rmsk.bed
$ bedops --merge rmsk.bed > rmsk.mergedRegions.bed
$ bedops --difference myRegions.bed rmsk.mergedRegions.bed > myRegions.masked.bed
$ bed2faidx.pl --options... < myRegions.masked.bed > myRegions.masked.fa

Etc.