Questions about the new version of DMFold

This forum is shown on the index page along with all topics.

Moderator: robpearc

Gary
Posts: 3
Joined: Sat Feb 24, 2024 5:59 am

Questions about the new version of DMFold

Post by Gary »

Hi, I was recently trying to predict protein complex using DMFold, and when I went to Step 2: Run_DeepMSA2.py, found that in the AF2 ranking step, there are some msa af2 predictions run templates_benchmark.py, such as aMSA, tthis should be the template search?
I0228 09:50:03.548033 46951213820416 templates_benchmark.py:883] Using precomputed obsolete pdbs /DeepMSA2/database/pdb_mmcif/obsolete.dat.
I0228 09:50:03.885781 46951213820416 tpu_client.py:54] Starting the local TPU driver.
I0228 09:50:03.886112 46951213820416 xla_bridge.py:212] Unable to initialize backend 'tpu_driver': Not found: Unable to find driver in registry given worker: local://
I0228 09:50:04.051980 46951213820416 xla_bridge.py:212] Unable to initialize backend 'tpu': Invalid argument: TpuPlatform is not available.
I0228 09:50:09.956907 46951213820416 run_alphafold_msa_benchmark.py:262] Have 5 models: ['model_1', 'model_2', 'model_3', 'model_4', 'model_5']
I0228 09:50:09.957167 46951213820416 run_alphafold_msa_benchmark.py:275] Using random seed 1822179567781964487 for the data pipeline
I0228 09:50:10.666351 46951213820416 hhsearch_benchmark.py:80] Launching subprocess "/DeepMSA2/anaconda3/envs/alphafold2nondocker/bin/hhsearch -i /tmp/tmpg1sihrja/query.a3m -o /tmp/tmpg1sihrja/output.hhr -maxseq 1000000 -z 100000 -Z 100000 -b 100000 -B 100000 -d /DeepMSA2/database/pdb70/pdb70"
I0228 09:50:10.773615 46951213820416 utils.py:36] Started HHsearch query
Is this normal?

In fact, I later found that parameter
alphafold_seqcut
affects whether template retrieval is performed. I would like to ask, if I only evaluate the quality of msa, do I not need to perform template retrieval, then I can modify alphafold_seqcut=0.0?
jlspzw
Posts: 245
Joined: Tue May 04, 2021 5:04 pm

Re: Questions about the new version of DMFold

Post by jlspzw »

Dear Gary,

AlphaFold2 default pipeline indeed uses HHsearch to detect the template which will be one of the inputs, so it is correct for modeling.

You are also correct if you do not want to use a template, you can set the seqcut parameter as 0, then the template features will be set as NA in the AF2. If you want to purely check the MSA quality. But for modeling, keeping the template will be OK.

Best Regards
Wei Zheng
Post Reply