Generate_msa --start-query for custom file

[rostro36]

Dear EvoDiff authors,
Thank you for making the code to EvoDiff available and showing the many facets of single sequence and MSA generation.
I had a closer look at generating MSAs from the query sequence and made some adjustments that I would like to contribute.

Changes

• README.md
  • wrong filename, larger description and added support for input file and output folder for generate_msa.py
• evodiff/data.py
  • Added flag to use special openfold treatment or more simple behaviour for custom a3m
  • Added support for single file instead of whole folder
• evodiff/generate_msa.py
  • Added output_fpath argument for output folder
  • Added support for custom input path
  • Again support for special openfold treatment or custom a3m
• data/openfold/test.a3m
  • Some test a3m

I tried to keep the code as similar as possible, except for more use of pathlib.Path. The defaults should still be the same and give the same output.

Tests

I ran it with the default parameters on my system for --start-query, but changed the input, output and batch size as I have not downloaded the whole OpenFold database. All tests gave me an output and I could see the different steps of diffusion ending in an MSA that was expectedly behind the results of the experiments in the pre-print, for which the models were built and much effort went in.

Please let me know about any feedback and changes that are still needed :)

[rostro36]

@microsoft-github-policy-service agree