single sequence prediction

[ShuhaoGithub]

Describe the bug

I am doing single sequence prediction with dummy msa of only the query sequence. I followed the instruction but got error that indicates use_msas, use_main_msas, use_paired_msas are not permitted field in the json.

To Reproduce

Command:

run_openfold predict --query_json=xxx.json

Query:
{
"queries": {
"xxx": {
"chains": [
{
"molecule_type": "protein",
"chain_ids": "A",
"sequence": "QVQLVESGGGLVQAGGSLRLSCAASGSISSISTMGWYRQAPGKERELVAAITSGGSTNYADSVKGRFTISRDNAKNTVYLQMNSLKPEDTAVYYCNFKYYSGSYFYKSEYDYWGKGTPVTVSS",
"main_msa_file_paths": "xxx.a3m"
}
]
}
}
}

A3M file: xxx.a3m
">query
QVQLVESGGGLVQAGGSLRLSCAASGSISSISTMGWYRQAPGKERELVAAITSGGSTNYADSVKGRFTISRDNAKNTVYLQMNSLKPEDTAVYYCNFKYYSGSYFYKSEYDYWGKGTPVTVSS"

Expected behavior

queries.xxxchains.0.use_msas
Extra inputs are not permitted [type=extra_forbidden, input_value=True, input_type=bool]
For further information visit https://errors.pydantic.dev/2.12/v/extra_forbidden
queries.xxxchains.0.use_main_msas
Extra inputs are not permitted [type=extra_forbidden, input_value=True, input_type=bool]
For further information visit https://errors.pydantic.dev/2.12/v/extra_forbidden
queries.xxx.chains.0.use_paired_msas
Extra inputs are not permitted [type=extra_forbidden, input_value=False, input_type=bool]
For further information visit https://errors.pydantic.dev/2.12/v/extra_forbidden

Configuration (please complete the following information):

• OS: Ubuntu 20.04
• GPU used and memory available: L40S 46068MiB
• Python version (if relevant): 3.13
• Installation path: mamba

[jnwei]

Hi @ShuhaoGithub

Thank you for starting to look into single sequence prediction!

I'm happy to report that we are actively working on adding better support for MSA-free / single sequence prediction. It should hopefully be complete within ~2 weeks.

In the meantime, I took a closer look at your example and your approach of providing a dummy a3m file with just the query sequence. I was able to run the example and get a better prediction compared to providing no MSAs by adding two small changes

• Provide the following MSA settings in a runner.yml file. The name xxx should be changed to match whatever the a3m file is named. More information about these settings can be found in our guide for using pre-computed MSAs for inference

# single_msa_runner.yml
dataset_config_kwargs:
  msa:
    max_seq_counts:
      xxx: 1 
    aln_order:
      - xxx 

• Change the prediction command to specify no msa server and use the above runner yaml, e.g.

run_openfold predict --query_json=xxx.json \
--use_msa_server=False \
--output_dir=test_xxx_single \
--runner_yaml=single_msa_runner.yml

The current query.json that you provided should work.

---

Here is a sample of the predictions:

Without using single query .a3m file in prediction (avg plddt = 28.5)

Using single query a3m file in prediction (avg plddt = 32.4)

[ShuhaoGithub]

Thank you very much for the swift reply. This is super helpful!