Handling token size expectations with paired MSAs and ligands

[BucherMichael]

I created unpaired and paired MSAs using jackhmmer searches and created two separate MSA contexts out of them. However, when you combine protein complex modeling with a ligand, the AllAtomStructureContext determines the token size from the number of protein residues plus the ligand atoms. In my hands, this produced a "BeartypeCallHintParamViolation" since the token size for MSAContext does not account for ligand tokens. I solved this by padding my paired MSA to the expected token size but most likely there is a better way to solve this?! Would be very happy for input! Many thanks!

[arogozhnikov]

I solved this by padding my paired MSA

this is correct approach (assuming that ligands go last in input).

[jackdent]

We'll have code for MSA loading released soon (you can follow along with progress here: #109)

[BucherMichael]

Fantastic, many thanks for the fast response and the great work!