singularity/apptainer compatibility

[seanrjohnson]

It would be great to be able to run protenix from a singularity/apptainer image. I tried the one below, but it didn't work. It seems to be trying to write cache database files to a directory that isn't writable in the apptainer file system. Maybe it would be possible to make the cache directory specifiable on the command line?

protenix.def

Bootstrap: docker
From: condaforge/miniforge3:24.9.2-0

%environment
export PATH="/opt/conda/bin:$PATH"

%post

conda install -c pytorch -c nvidia -c conda-forge  compilers python=3.11 pytorch pytorch-cuda=12.4 cuda-nvcc
pip install protenix -U

 apptainer build protenix.sif protenix.def

apptainer exec --nv protenix.sif protenix predict --input protenix_config.json --out_dir protenix_prediction

output:

Try to find the ccd cache data in the code directory for inference.
[2025-06-02 16:55:49,344] [INFO] [real_accelerator.py:254:get_accelerator] Setting ds_accelerator to cuda (auto detect)
2025-06-02 16:55:52,518 [/opt/conda/lib/python3.11/site-packages/runner/batch_inference.py:255] INFO runner.batch_inference: run infer with input=protenix_config2.json, out_dir=protenix_2_oligo_2_prots, cycle=10, step=200, sample=5, use_msa_server=False
2025-06-02 16:55:52,519 [/opt/conda/lib/python3.11/site-packages/runner/batch_inference.py:212] INFO runner.batch_inference: will infer with 1 jsons
Traceback (most recent call last):
  File "/opt/conda/bin/protenix", line 8, in <module>
    sys.exit(protenix_cli())
             ^^^^^^^^^^^^^^
  File "/opt/conda/lib/python3.11/site-packages/click/core.py", line 1442, in __call__
    return self.main(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/conda/lib/python3.11/site-packages/click/core.py", line 1363, in main
    rv = self.invoke(ctx)
         ^^^^^^^^^^^^^^^^
  File "/opt/conda/lib/python3.11/site-packages/click/core.py", line 1830, in invoke
    return _process_result(sub_ctx.command.invoke(sub_ctx))
                           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/conda/lib/python3.11/site-packages/click/core.py", line 1226, in invoke
    return ctx.invoke(self.callback, **ctx.params)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/conda/lib/python3.11/site-packages/click/core.py", line 794, in invoke
    return callback(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/conda/lib/python3.11/site-packages/runner/batch_inference.py", line 259, in predict
    inference_jsons(
  File "/opt/conda/lib/python3.11/site-packages/runner/batch_inference.py", line 219, in inference_jsons
    runner = get_default_runner(seeds, n_cycle, n_step, n_sample)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/conda/lib/python3.11/site-packages/runner/batch_inference.py", line 181, in get_default_runner
    download_infercence_cache(configs, model_version="v0.5.0")
  File "/opt/conda/lib/python3.11/site-packages/runner/inference.py", line 170, in download_infercence_cache
    os.makedirs(os.path.dirname(cur_cache_fpath), exist_ok=True)
  File "<frozen os>", line 215, in makedirs
  File "<frozen os>", line 225, in makedirs
OSError: [Errno 30] Read-only file system: '/opt/conda/lib/python3.11/site-packages/release_data'

[zhangyuxuann]

@seanrjohnson Hi，can you download the ccd cache and model checkpoint to a directories you have permission to.

/xxx/path/to/your/dir
 - components.v20240608.cif
 - components.v20240608.cif.rdkit_mol.pkl
 - model_v0.5.0.pt

wget -P /xxx/path/to/your/dir/ https://af3-dev.tos-cn-beijing.volces.com/release_model/model_v0.5.0.pt
wget -P /xxx/path/to/your/dir/ https://af3-dev.tos-cn-beijing.volces.com/release_data/components.v20240608.cif
wget -P /xxx/path/to/your/dir/ https://af3-dev.tos-cn-beijing.volces.com/release_data/components.v20240608.cif.rdkit_mol.pkl

than modify the ccd cache and model checkpoint path.

[seanrjohnson]

Thanks!

I didn't want to change the source code (because I don't want to fork the repo and have to maintain my own fork). I found a solution that worked though. I just run the test example in my container creation script. So it downloads the ccd and checkpoint before the image becomes read only.

protenix.def

Bootstrap: docker
From: condaforge/miniforge3:24.9.2-0

%environment
export PATH="/opt/conda/bin:$PATH"

%post

conda install -c pytorch -c nvidia -c conda-forge  compilers python=3.11 pytorch pytorch-cuda=12.4 cuda-nvcc
git clone https://github.com/bytedance/Protenix.git
cd Protenix
pip install -e .
protenix predict --input examples/example.json --out_dir  ./output --seeds 101 --cycle 1 --step 1 --sample 1

Then have to build with --nv for it to not be super slow:
apptainer build --nv protenix.sif protenix.def

When I do it that way, the container seems to work as expected.

[zhangyuxuann]

@seanrjohnson Thanks for your sharing.