alphapulldown 2.0.1 - create_individual_features.py with pre-computed mmseqs MSAs

[wresch]

When i generate MSAs locally with colabfold i end up with a folder like

$ ls -lh pulldown_cf_msas
-rw-r--r-- 1 user group 1.9M Jan 31 17:06 O76094.a3m
-rw-r--r-- 1 user group 9.5M Jan 31 17:06 P08240.a3m
-rw-r--r-- 1 user group 935K Jan 31 17:06 P09132.a3m
-rw-r--r-- 1 user group 2.0M Jan 31 17:06 P22090.a3m

after renaming the sequences.

then i try to use the existing MSAs with

create_individual_features.py \
  --fasta_paths=bait.fasta,candidates.fasta \
  --output_dir=pulldown_cf_msas \
  --use_precomputed_msas=True \
  --use_mmseqs2 \
  --max_template_date=2023-01-01 \
  --skip_existing=False

and unlike in older versions (at least in 0.30.7 and 1.0.4) this now hits the colabfold API and re-generates the MSA - at least it does a request and fails on systems without direct access to the internet.

AlphaPulldown/alphapulldown/scripts/create_individual_features.py

Line 306 in e4988f5

# Create features

looks like use_precomputed_msas is not passed to the feature generation function if mmseqs is used. If i don't specify --use_mmseqs2 then it regenerates the MSA with the alphafold pipeline b/c it expects a subdirectory containing the MSAs generated by that pipeline instead of the single .a3m file.

Some of this is supposed to be used for large-ish workflows so we want to avoid hitting the API to much and we do eventually get blocked.

The documentation suggests this should still be possible (https://github.com/KosinskiLab/AlphaPulldown?tab=readme-ov-file#run-mmseqs2-locally). Am i missing something?

[christophista]

Hello
Thank you for opening the issue. I have the same problem.
The pipeline for running

1. colabfold_search,
2. rename and then
3. create_individual_features.py

also accesses the MSA server and performs the alignment there again.

Thank you for your help in advance!

[sonaida]

Hello,
I have the same problem trying locally run MMseqs2.
Two log files of create_individual_features.py. The same code, the same alignment file.

Before: (AlphaPulldown/1.0.4) - works fine

I0205 18:09:59.794106 140456668571456 templates.py:857] Using precomputed obsolete pdbs /opt/gensoft/data/AlphaPulldown/1.0.4/pdb_mmcif/obsolete.dat.
I0205 18:09:59.886897 140456668571456 create_individual_features.py:191] running mmseq now
I0205 18:09:59.887056 140456668571456 objects.py:189] You chose to calculate MSA with mmseq2.
Please also cite: Mirdita M, Schütze K, Moriwaki Y, Heo L, Ovchinnikov S and Steinegger M. ColabFold: Making protein folding accessible to all. Nature Methods (2022) doi: 10.1038/s41592-022-01488-1
I0205 18:09:59.887385 140456668571456 objects.py:198] looking for possible precomputed a3m at Colab_msas_bait-gp03/BAB65153.1.a3m
I0205 18:09:59.887429 140456668571456 objects.py:200] input is Colab_msas_bait-gp03/BAB65153.1.a3m
I0205 18:09:59.890167 140456668571456 objects.py:203] Finished parsing the precalculated a3m_file
Now will search for template
I0205 18:09:59.898038 140456668571456 objects.py:242] will search for templates in local template database
I0205 18:09:59.899698 140456668571456 hmmbuild.py:121] Launching subprocess ['/opt/conda/envs/AlphaPulldown/bin/hmmbuild', '--amino', '/local/scratch/tmp/tmpi9vuoqyf/output.hmm', '/local/scratch/tmp/tmpi9vuoqyf/query.msa']
I0205 18:09:59.901239 140456668571456 utils.py:36] Started hmmbuild query
I0205 18:09:59.981078 140456668571456 hmmbuild.py:128] hmmbuild stdout:

Now: (AlphaPulldown/2.0.1)

I0205 17:51:08.065152 139704594203264 create_individual_features.py:308] Running MMseqs2 for feature generation...
I0205 17:51:08.067650 139704594203264 objects.py:188] You chose to calculate MSA with mmseq2.
Please also cite: Mirdita M, Schütze K, Moriwaki Y, Heo L, Ovchinnikov S and Steinegger M. ColabFold: Making protein folding accessible to all. Nature Methods (2022) doi: 10.1038/s41592-022-01488-1
0%| | 0/150 [elapsed: 00:00 remaining: ?]
SUBMIT: 0%| | 0/150 [elapsed: 00:00 remaining: ?]
PENDING: 0%| | 0/150 [elapsed: 00:00 remaining: ?]E0205 17:51:09.542120 139704594203264 colabfold.py:219] Sleeping for 6s. Reason: PENDING
PENDING: 0%| | 0/150 [elapsed: 00:07 remaining: ?]E0205 17:51:16.126149 139704594203264 colabfold.py:219] Sleeping for 9s. Reason: PENDING
RUNNING: 0%| | 0/150 [elapsed: 00:16 remaining: ?]
RUNNING: 6%|▌ | 9/150 [elapsed: 00:16 remaining: 04:22]E0205 17:51:25.705300 139704594203264 colabfold.py:219] Sleeping for 9s. Reason: RUNNING
COMPLETE: 6%|▌ | 9/150 [elapsed: 00:26 remaining: 04:22]
COMPLETE: 100%|██████████| 150/150 [elapsed: 00:26 remaining: 00:00]
COMPLETE: 100%|██████████| 150/150 [elapsed: 00:27 remaining: 00:00]
I0205 17:51:36.024016 139704594203264 batch.py:799] Sequence 0 found no templates

[Jun2BCR]

Hello,
Thank you so much for bringing up this question. I am experiencing the same issue. It will be highly appreciated if someone can help.