SuCOS-pocket similarity calculation

[YoavShamir5]

Hello,

Thanks for making this innovative and super useful tool out there!

Given a single AF3 prediction of a protein-ligand complex, my goal is to find the PDB protein-ligand complex with the highest SuCOS-pocket similarity to it (or the top 10 most similar PDB complexes, for example). It seems like your code supports this, but I would appreciate any advice on how to do so. Also - I wonder if the runtime (calculating similarity against the entire PDB) is expected to be extremely long.

Thanks for your input!

[Ninjani]

We are working on making this available in an ongoing PR in PLINDER: plinder-org/plinder#105

It does take quite some resources as you'd need all of PLINDER downloaded and to run Foldseek against the entire PDB, working on trying to reduce requirements here. In terms of runtime it's not so bad - takes ~20min to run the Foldseek search and then for most complexes 2-30min to calculate the SuCOS pocket_qcov scores. For some it takes way longer (>1day) just because there are a lot of Foldseek hits and we would need to calculate the SuCOS scores against all the hits to get the top ones.