search

google-deepmind / alphafold

Open source code for AlphaFold.
Apache License 2.0
12.35k stars 2.21k forks source link

fine-tuning trained models and associated parameters for multimers #237

Closed arashnh11 closed 2 years ago

arashnh11 commented 2 years ago

So in v2.0.0 there was a flag called --CASP14 which had fine-tuned/optimized parameters for the challenging testcases of monomers.

We are now evaluating v2.1.1 for an octamer and so far getting RMSD ~ 14 for alignment with one of the models (rank_3 in this case, not rank_0, surprisingly). Are there any suggestions about how to fine-tune the training models and associated parameters based on use cases? Or how to even get started in that route?

We are happy to share all our findings based on your suggestions.

sbodenstein commented 2 years ago

Sorry about that, unfortunately this repo only provides support for evaluating the pre-trained AlphaFold models, not for finetuning on new data. So you would need to implement a full training pipeline.