search

hcgasser / CAPE

2 stars 1 forks source link

MHC Class 1 position weight matrices generation error #5

Closed diossynvax closed 6 months ago

diossynvax commented 6 months ago

Hi

I seem to have a problem with the MHC-I_rank_peptides.py script. The script generates the random_peptides.txt file but then freezes at the next stage. I don't know if the issue here is memory related but I have attempted this on an ml.p3.2xlarge and an ml.p3.8xlarge instance with the same result. It appears that netMHCpan has installed correctly so I'm not quite sure what the issue could be. Could you make a suggestion as to what might be going wrong?

Output below.

MHC-I_rank_peptides.py --output ${PF}/data/input/immuno/mhc_1/MhcPredictorPwm --alleles ${MHC_Is} --peptides_per_length 1000000 2024-05-10 12:41:47: created directory: /CAPE/data/input/immuno/mhc_1/MhcPredictorPwm Generate new random_peptides Generate random peptides: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 3/3 [01:23<00:00, 27.73s/it] HLA-A*02:01 - rank: 0%| | 0/6 [00:00<?, ?it/s]2024-05-10 12:43:20: Init <class 'kit.bioinf.immuno.mhc_1._netmhcpan.Mhc1PredictorNetMhcPan'> MHC-Class1 predictor

hcgasser commented 6 months ago

This steps generates a first order approximation for netmhcpan predictions - the less peptides are used for this the less precise this will be. Have you tried with less peptides? (e.g. --peptides_per_length 1000). If this works it will likely be some hardware or time constraint. Tbh... this takes A LOT OF TIME (which is actually the reason why we produce the first order approximation for the behaviour of netmhcpan). So you might just have to wait longer when using more peptides.

diossynvax commented 6 months ago

I've tried a few more tests and I think it's just a memory issue. I'll adjust accordingly. Thanks for the quick response.