r-krishna
commented
6 months ago there is some stochasticity because of MSA clustering, I would expect that if you ran the model multiple times you might get something closer to the example in the paper.
Re histidine: that is weird I have not seen that before. I will look into it
Since I can run Predicting Protein Monomers example well, I tried Predicting Covalently Modified Proteins example verbatim (i.e., reused given input files and input config).
Unfortunately, I see a jumbled backbone as a red box here
I see that glycan part is modeled as HETATM at the end of result file (i.e., covalent/7s69_A.pdb),
,
and canonical peptide bond backbone atoms in most pdb files (i.e., N, CA, C, O) are not specified (rather they are specified as just N,C,O).
Are these given input files (i.e., examples/protein/7s69_A.fasta and examples/small_molecule/7s69_glycan.sdf) and config file are same that generated RFAA paper Fig. 3-F ?
Or are results of RFAA stochastic (i.e., sometimes good, sometimes bad)?
Additionally, modelled histidine structure is not linear as a red box here
. I've acquired non-linear histidine structure from my other personal (i.e., not RFAA official tutorial) RFAA run as well.
Overall protein structure comparison between native 7s69_A and RFAA result looks ok, but I couldn't reproduce glycan orientation/conformation as paper showed.