Repeatmasker run with crossmatch

[gen2prot]

Hello All,

I am running repeatmasker on a 2.1G genome that is assembled into 10 psuedomolecules. I am using the following parameter

-engine crossmatch -pa 32 -nolow -norna -cutoff 250 -gff

I am running the job on the lsf cluster, but even after allocating sufficient memory and 32 processors per chromosome, its been running for 72 hours. I am using a database containing 3305 repeat sequences. Is there a method to increase the speed of the run?

Please let me know.
Thank you

[Brian Bushnell]

Depending on exactly what you are trying to do, you could certainly achieve the masking a lot faster (in a few minutes) using BBDuk instead via kmer-based masking. For example:

bbduk.sh in=genome.fa out=masked.fa k=31 ref=repeats.fa kmask=N

That will mask everywhere that shares 31-mers with the reference. It also supports mismatches and sliding-window entropy masking. bbmask.sh is slightly more complete and can mask short repeats or mask a genome from an aligned sam file, and is also very fast.

[gen2prot]

Hi Brian,

I would like to use the masked genome for repeat annotation. Does use of a kmer based masking tool lead to over masking? I would rather use a signature based masking tool like LTR_Struc, LTR_harvest in combination with Repeatmasker. What do you think?

Thanks

[Brian Bushnell]

That's an excellent question. I really don't know if it would be more or less sensitive, or better or worse, as I have l no direct experience with annotation. It might mask little regions in the middles of genes, for example (though you can circumvent that by using a very large K). Possibly a better solution would be to use a long read aligner and mask the regions the repeats align to. The big problem here is that I still don't understand the rationale for masking prior to annotation, but it seems to be due to some weakness in annotation algorithms - in other words, it's algorithm-specific rather than procedure-specific. If that's correct, that makes it impossible to determine the best method of masking (or whether masking is even beneficial) without a very deep understanding of the specific annotation algorithm.

If you have some way of empirically determining the quality of annotation, though, it's certainly cheap to try alternative masking methods for comparison.