search

sokrypton / ColabFold

Making Protein folding accessible to all!
MIT License
1.85k stars 471 forks source link

Does multimer prediction takes into account experimental structure published after training? #458

Open remybonnav opened 1 year ago

remybonnav commented 1 year ago

I am trying to get some predictions of protein-protein complexes. As "positive" control, I use prediction for a complex for which the cryoEM structure was solved and published in second half of 2022. When I use colab fold multimer (from ChimeraX software), does the prediction of the structure is using somehow the published structure or is it only relying on the model trained (that does not contain the published structure because it came out after alpha fold multimer)?

milot-mirdita commented 1 year ago

The template database used currently in by the ColabFold server is from March 2022. I am in the progress of updating the template db, but that will still take a bit.

The knowledge cutoff for multimer-v3 is Sept 30, 2021 (https://twitter.com/sokrypton/status/1620808280238084097).