Speed enhancement about ssw HOT 8 CLOSED

I just pushed a commit that provided a considerable speed up of convert_sequence_to_ints, but the the speed up of alignment_report is much trickier. One problem I know about is that the header of alignment_report essentially calls alignment three times (first for the match count, second for the mismatch count, and third for the report itself). The problem is that the CIGAR string does not differentiate between matched and mismatched bases, only that bases align. The reason I'm calling upper so much is to preserve the upper/lower case of input sequences in the subsequent report. The getseq() closure is a convenience, and really doesn't lend to itself to the inefficiencies you are experiencing. I've thought about caching the results, but I also wanted to avoid memory bloat. A quick win to try is setting the header=False on the alignment_report, or just using the alignment property, and formatting your own version of the content.

from ssw.

Great, Thanks!

I fixed a speedup to alignment(). I have only profiled one data set, where it seems to give about 2x speedup.

ncalls  tottime  percall  cumtime  percall filename:lineno(function)
88540   49.172    0.001   82.287    0.001 sswobj.py:185(alignment)

Compare to the same profile row in previous post:
88412 52.584 0.001 158.666 0.002 sswobj.py:149(alignment)

I noticed slightly more calls to the new run but that should be change in the parameters in my software, i.e., not related to the alignment function itself (if does not call itself) . I did assert on the output of yours and mine alignment function and it passed all of my 3700 alignments (biological data), hopefully it will pass your unittests as well. Could you consider using this function instead? The function is found here:
https://gist.github.com/ksahlin/76a16f0b6fa19988b1db

from ssw.

correction: should be "it does not call itself"

from ssw.

Hi again, I did a further optimization in the alignment() function, specifically of the match string parsing as:

match_seq = ''.join(['|' if r_base.upper() == q_base.upper() else '*' for (r_base, q_base) in zip(ref_piece, query_peace)])

see gist https://gist.github.com/ksahlin/3b740b78d1d80cbd20be
new benchmark:

    88540   35.453    0.000   61.895    0.001 sswobj.py:185(alignment)

from ssw.

Hi again,

could you consider including the function i attached in the gist in your repo?

from ssw.

@ksahlin why not fork it and submit a PR?

from ssw.

@Chris7 Ok! Done in #5

from ssw.

Thanks for this PR, this has been integrated.

from ssw.