openfold
openfold copied to clipboard
Published
20 hours ago •
aqlaboratory
aqlaboratory
The procedure of making the self-distillation data.
I'm curious about .pdb files your kindly provided for self-distillation. Were they predicted using official AF2 weights or the openfold initial training weights trained from scratch?
Oh, I found this. Still curious about which model was used. I checked some cases, those structures seem not precisely the same as in the Alphafold DB, especially some regions having low plddt.
Also, a following up discussion, if the sequences were removed greedily, should we be concerned about enriching too many 'orphan' sequences? I see you kept sequences with more than 50 MSA only, I guess there are many sequences that were dropped.