When I try to model a protein from the PDB (4WYH) I follow the abintio protocol correctly and generate fragments from Robetta. Yet I do not get a funnel shaped plot after 25,000 decoys.
Though it’s sequence clearly folds into this particular structure (as proved by the crystal structure) why can’t abinitio performs a successful simulation and result in a funnel shaped plot?
Should I change the abinitio protocol from the tutorial, or run it multiple times, or generate more fragments to be able to computationally model the true shape of this protein and get a funnel plot?
to be blunt -- this is because structure prediction is an unsolved problem! while abinitio is relatively good at solving some structures, it isn't able to solve them all!
Generally the field has strayed away from using abinitio exclusively, and has added iterative rounds of rosetta hybridize as described here https://www.ncbi.nlm.nih.gov/pubmed/26677056 I'm aware of many scenarios where a seemingly failed abinitio run can be signifigantly improved via 4-10 rounds of iterativehybridize. Also described in that paper is using predicted contacts to drive abinitio folding, which is sometihng you may be interested in using (sergeys gremlin protocol)