facebookresearch
Folded structure from `esm-fold` has overlapping atoms
Bug description I installed esm using the conda environment I Folded many proteins using the esm-fold command, however I have found some proteins with physically impossible structure.
Reproduction steps
input fasta
test.txt
output pdb
test_pdb.txt
clipped out first 9 residues for image model
Trying with the whole PDB will only crash your viewer
Expected behavior Expected 3D structure with physically possible coordinates
Logs Please paste the command line output: 23/05/11 02:23:45 | INFO | root | Predicted structure for testAA with length 1046, pLDDT 29.2, pTM 0.999 in 126.8s. 31536 / 32805 completed
Additional context I do not udnerstand how to specify model used by ESM-fold command (maybe there are no other options for this command?). I'm supposing that this could be due to trying to fold a protein with > 1000 AA, but the limits of the folder are not very explicit. Like maybe there's a 400 AA limit, but that's mentioned and then quickly glazed over.
I have a similar problem, where the predicted structure has missing residues.
