[wgs-assembler-users] pbdagcon on Drosophila reads

[A. B. C. <ber...@gm...>]

Hi,
I noticed that while the Drosophila melanogaster   MHAP assembly  is very
good in general, it has many gaps in single-copy Y-linked genes. I guess
that this is caused by low coverage: the DNA came from males,  and was
assembled at 25x, which leaves the Y genes at 12.5x (theoretically).
Furthermore, it seems that Y-linked reads are being lost during the first
correction step (done by falcon-sense; I checked the uncorrected and the
corrected reads).

 I am trying to fix these problems by increasing the coverage of the
corrected reads used in the "post-correction"  steps (by adding
assembleCoverage=40  in the spec file ; instead of the default 25x) , and
by forcing the use of pbdagcon instead of falcon-sense (by adding pbcns=1
in the spec file).  The assembly with 40x and falcon-sense worked fine ,
but when I tried 40x with pbdagcon , the run seems to be abnormally slow.
Specifically, the machine I used is a Dell with 24 processors / 144 Gb RAM,
 and after 9 days running it was still processing the first two partitions
of runPartition.sh

# /home3/users/bernardo/drosophila//tempdros10/runPartition.sh 1
# /home3/users/bernardo/drosophila//tempdros10/runPartition.sh 2

I checked the runPartition.sh script, and it seems to use only 8 threads
(instead of 24):

cat  /home3/users/bernardo/drosophila//tempdros10/runPartition.sh

  $bin/outputLayout \
      -L \
      -e 0.35 -M 1500 \
      -i /home3/users/bernardo/drosophila//tempdros10/asm \
      -o /home3/users/bernardo/drosophila//tempdros10/asm \
      -p $jobid \
      -l 500 \
       \
      -P \
      -G /home3/users/bernardo/drosophila//tempdros10/asm.gkpStore \
 2> /home3/users/bernardo/drosophila//tempdros10/$jobid.lay.err |
$bin/convertToPBCNS -consensus pbdagcon -path
/home3/users/bernardo/programs/wgs-8.3rc2/Linux-amd64/bin/ -output
/home3/users/bernardo/drosophila//tempdros10/$jobid.fasta -prefix
/home3/users/bernardo/drosophila//tempdros10/$jobid.tmp -length 500
-coverage 4 -threads 8 >
/home3/users/bernardo/drosophila//tempdros10/$jobid.err 2>&1 && touch

In this particular run  I have not specified cnsConcurrency or
consensusConcurrency in the spec file (so the PBcR choose the values; I
only set threads=20 ),  but in another run I added cnsConcurrency=20
consensusConcurrency=20
to the spec file, and again in 10 days it processed only 3 of the 200
partitions.

I tried before the ecoli 30x and the yeast data, and both worked fine with
pbdagcon (although slower than falcon-sense). Are there some limitation to
use pbdagcon with higher coverage data? Is the -threads 8  option of  the
convertToPBCNS program correct?

Thanks,
Bernardo









A. Bernardo Carvalho

Departamento de Genética
Universidade Federal do Rio de Janeiro