Segmentation fault on scan_all function about gimmemotifs HOT 5 OPEN

Thank you for sharing this! I was able to reproduce the error with test data, but it depended on the input fasta!

from gimmemotifs.motif import Motif, read_motifs
from gimmemotifs.fasta import Fasta


jaspar_motif_file="data/motif_databases/JASPAR2022_vertebrates.pfm"
motifs = read_motifs(jaspar_motif_file)
print(len([m for m in motifs if m.id.startswith("MA0046.2")]))  # prints 1

m = [m for m in motifs if m.id.startswith("MA0046.2")][0]
print(m.id)  # prints 'MA0046.2_MA0046.2.HNF1A'

ff = "test/data/scan/genome/scan_test.fa"  # segfault
ff = "test/data/genomes/hg38sample.fa"  # no segfault

f = Fasta(ff)
m.scan_all(f)

The file with proper output contained one chromosome with zero hits, so I'm not sure what causes this. @simonvh, could you have a look if you have the time? This is happening inside the C function pfmscan.

from gimmemotifs.

Hello,

I believe, I have identified problem - when a sequence in FASTA file is shorter then motif, I believe that is when segmentation fault happens.

For example if I have something like this in FASTA file (usually incomplete CDS):

ENSG00000146733|ENST00000416592
ATGGTCT

And consensus is for example (https://jaspar.genereg.net/matrix/MA0114.4/):
'nnCAAAGTCCAnn'

Similar happens on the other short sequences.

from gimmemotifs.

I only got a segfault with a long sequence, but not with a short sequence!
Perhaps I'm missing something though, so I made a PRs that might fix either cases: #303 and #304

If you could test these, or provide me with your input data, that would be wonderful :)

from gimmemotifs.

You can download file I have used here: https://filebin.net/sxmgfi99b5rnzb1y

from gimmemotifs.

Thanks!

I used your example code, with your mart file, and this MA0046.2 motif (I guess its the same you had?) but did not get a segfault!

I'm using a beefy PC, so my theory is that I have a bigger CPU cache (stack, I think?). If that is the case, #303 / #305 should fix this issue. Let me know what you think @simonvh :)

from gimmemotifs.