Paired/Unpaired MSA switch

[Qrouger]

The paired MSA is sliced and placed in the unpaired MSA, while the paired MSA field is set to an empty string (""). This causes an issue with AlphaFold 3, because in this configuration AF3 ignores inter-protein paired MSAs and therefore does not use co-evolutionary information between proteins.

[DimaMolod]

The paired MSA is sliced and placed in the unpaired MSA, while the paired MSA field is set to an empty string (""). This causes an issue with AlphaFold 3, because in this configuration, AF3 ignores inter-protein paired MSAs and therefore does not use co-evolutionary information between proteins.

Hi Quentin! That's exactly my thoughts a couple of months ago :-)
I think you misunderstand what AF3 does with 'paired_msa' and 'unpaired_msa' (which is not surprising because this is indeed very confusing). The hint here is to read 'paired_msa' as 'to_be_paired_msa', and 'unpaired_msa' as 'not_to_be_paired_msa' (see af3 documentation). If you put paired MSA as 'unpaired_msa' variable, AF3 doesn't do additional pairing based on species id and takes your MSA as is. This is exactly what we want when passing MultimericObject from AlphaPulldown, because MSAs are already paired at this moment by AF2.
There are flags in run_structure_prediction.py here that you can use to double-check after me.

Cheers
d