[BUG] MSA related features not allowed at the chain level.

[RHoejland]

Describe the bug

Currently, the documentation Input Docs states that msa behaviour can be controlled at the chain level.
However the msa related keys: "use_msas", "use_paired_msas", and "use_main_msas" are validated only at the Query level and not inherited by the chain class.
This results in a validation error when attempting to control msa behaviour at chain level. I.E. when any of the above keys are defined within a chain section, e.g. as the example input example shows.

To Reproduce
A minimal json to reproduce the issue, this is directly taken from the example linked previously.

{
    "queries": {
        "query_1": {
            "chains": [
                {
                    "molecule_type": "protein",
                    "chain_ids": "A",
                    "sequence": "PVLSCGEWQCL",
                    "use_msas": true,
                    "use_main_msas": true,
                    "use_paired_msas": true
                }
            ]
        }
    }
}

No custom runners were used; and the simple command:
run_openfold predict --query-json=msa_error.json
was used.
Expected behavior
The msa settings should be parsed and applied to the specific chain, following which inference would go on as normal.

Stack trace

Traceback (most recent call last):
  File "/usr/local/bin/run_openfold", line 7, in <module>
    sys.exit(cli())
             ~~~^^
  File "/usr/local/lib/python3.13/site-packages/click/core.py", line 1485, in __call__
    return self.main(*args, **kwargs)
           ~~~~~~~~~^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.13/site-packages/click/core.py", line 1406, in main
    rv = self.invoke(ctx)
  File "/usr/local/lib/python3.13/site-packages/click/core.py", line 1873, in invoke
    return _process_result(sub_ctx.command.invoke(sub_ctx))
                           ~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^
  File "/usr/local/lib/python3.13/site-packages/click/core.py", line 1269, in invoke
    return ctx.invoke(self.callback, **ctx.params)
           ~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.13/site-packages/click/core.py", line 824, in invoke
    return callback(*args, **kwargs)
  File "/usr/local/lib/python3.13/site-packages/openfold3/run_openfold.py", line 191, in predict
    query_set = InferenceQuerySet.from_json(query_json)
  File "/usr/local/lib/python3.13/site-packages/openfold3/projects/of3_all_atom/config/inference_query_format.py", line 98, in from_json
    return cls.model_validate_json(data)
           ~~~~~~~~~~~~~~~~~~~~~~~^^^^^^
  File "/usr/local/lib/python3.13/site-packages/pydantic/main.py", line 766, in model_validate_json
    return cls.__pydantic_validator__.validate_json(
           ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^
        json_data, strict=strict, extra=extra, context=context, by_alias=by_alias, by_name=by_name
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    )
    ^
pydantic_core._pydantic_core.ValidationError: 3 validation errors for InferenceQuerySet
queries.query_1.chains.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.query_1.chains.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.query_1.chains.0.use_paired_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

Configuration (please complete the following information):

• OS: Ubuntu 22.04.5 LTS - Container base image: Docker: python:3.13-bookworm (Debian)
• GPU: RTX 4060 Ti 16GB (Higher memory GPU's also available, however this issue newer gets to allocate anything :) )
• Python version (if relevant): 3.13
• Installation path: Containerized with apptainer, package built from repo.

[jnwei]

Hi @RHoejland thank you for your interest in OpenFold3 and for reporting this error.

I'm afraid this the Input Format documentation is incorrect.

In our development, we opted not to support per-chain handling of MSAs, in favor of handling them at the query level. In other words for manual construction of the input jsons, msa fields (e.g. use_msas, use_paired_msas, use_main_msas) should be set at the query level. If that doesn't work for your use case, let us know and we can try to thelp.

[RHoejland]

Thanks for the swift reply, typically we would like to run predictions on heteromers with MSA limited to just a single chain. From my understanding of the documentation this would be possible by providing pre-computed MSA's where all but one of them are single-sequence MSA. (I.E. just the chains own sequence in the alignment file). Is this correctly understood?

If that is the case it should be fairly straight forward to generate the empty MSA's on my end.

Hi @RHoejland thank you for your interest in OpenFold3 and for reporting this error.

I'm afraid this the Input Format documentation is incorrect.

In our development, we opted not to support per-chain handling of MSAs, in favor of handling them at the query level. In other words for manual construction of the input jsons, msa fields (e.g. use_msas, use_paired_msas, use_main_msas) should be set at the query level. If that doesn't work for your use case, let us know and we can try to thelp.