On 19 Apr 2014, at 1:53 , Robert Bruccoleri <bruc@xxxxxxxxxxxxxxxxxxxxx> wrote:
I'm working on a bacteria which has a lot of apparently repeated sequences. The file in
the "_d_info" directory whose name ends in _readrepeats.lst is 340 MB long. The
repeat ratio tabulation in the log file shows that there are lot of reads containing
sequences repeated 100's of time relative to most of the genome.
100’s of times? You’re having some interesting bacteria there. If I had to
guess: a non lab strain with at least 3 MB and lots of transposons / insertion
elements. One or multiple phages may also be a possibility there.
[…]
My goal is to figure out what's repeated.
[…]
but it's clear the contigs are chimeras.
[…]
I haven't tried to optimize this manifest -- it gives me results that I can
interpret, but I wonder if there's a better solution out there.
We’re speaking about Illumina data, right? Then I would approach this from
another angle.
Use the read list in the debris file to extract the reads with repeats from
your whole set (maybe just take the reads marked MNRr, venturing into HAF7 and
HAF6 if you really want). Then treat these reads as if they were coming from
transcripts and assemble them in EST mode. Note: just having been reminded of
this in another thread on the list … switch off digital normalisation and the
masking of nasty repeats for this special case.
With that approach the normal security measures of MIRA for assuring data
quality are kept working, and repeats having differences only in single bases
are still kept apart, probably reducing the number of chimeras. As bonus, you
will be able to deduce the approximate number of copies for each repetitive
element in the genome by putting the contig coverage numbers of the “transcript
assembly” in relation to the average genome coverage you will have gotten from
the whole genome assembly. WARNING: this assumes the DNA of your organism has
been extracted in stationary phase. If this was done in exponential phase you
have a couple of other problems.
Hope that helps,
Bastien
PS: oh, and I suppose I do not need to point you at PacBio, do I?
