Repeated RosettaDock crashes

[ankimble]

I've been using RosettaDock 5.0 to determine a ligand ability to interact with a peptide. One of the ligands, phosophocholine (PC), appears to lead to normal docking procedures. Anything else that I tried consistently fails and sends out a crash log. I can't by eye determine the difference other than may of the other ligands are ring structures like inositol. I've even tried placing them in other starting locations and it still fails. I've got at least 3 different ring ligands that were all developed independently and I'm getting the same crashing. I'm attaching examples of the ones that worked and didn't work. I need help because this was specifically requested by a reviewer.

examples docking.zip

[roccomoretti]

For rosie-61191.VSEAYENLV_7_inositol, It looks like your ligand.pdb contains a residue labeled UNK. That's an "unknown ligand" by the wwPDB, and as such Rosetta doesn't have the ability to load it. It will just ignore the residue, leading to the "Cannot compute center of mass of zero residues!" error message you see. (The UNK residue was discarded, so there aren't any residues in the ligand partner.)

Additionally, it looks like the occupancy column of the ligand is 0.00 -- Rosetta will also discard any zero occupancy atoms, as atoms with no occupancy are usually poorly modeled. Even if it wasn't discarding the UNK residue, it would discard all the ligand atoms due to zero occupancy.

A third issue you'd run into is that your ligand does not have unique atom names. All of the atom names are just the element codes, which means there's multiple 'C' atoms in the ligand, which will confuse Rosetta.

But perhaps your biggest issue is that you're attempting to use a protein-protein docking protocol to dock a peptide with a small molecule ligand. While this may technically work (e.g. with phosophcholine), one of the big issues you'll run into with ROSIE is that the RosettaDock 5.0 application on ROSIE is set up assuming that you're only be using the standard amino acids. Rosetta needs a "params file" for each residue to tell it information about the chemistry of the residue (atom types, bonding, etc.). While it comes with params for most of the common residues (e.g. all the canonical amino acids and certain other compounds like phosophocholine), params for other compounds aren't present by default and have to be provided manually. Unfortunately, the ROSIE version of RosettaDock can't take these additional compounds.

You may have better success using the Ligand Docking application to do the docking. This is intended for protein/small molecule docking, with potentially arbitrary small molecules. It allows you to specify an arbitrary molecule as an SDF formatted file. This should work for things like inositol or phosophocholine, though it doesn't work so well if both partners are polypeptides. (Which is where RosettaDock and FlexPepDock come in.)

That said, the RosettaLigand ligand docking has only really been benchmarked for globular protein docking. I'm not sure how well it will perform with peptide interactions. (This is not really something that Rosetta has been benchmarked against.) -- One of the main assumption these docking programs (RosettaLigand, RosettaDock, FlexPepDock) make is that the "receptor" side of the complex is a well structured protein which already contains the binding pocket. There's some small internal flexibility of the receptor, but there really isn't any active sampling of degrees of freedom. As such, they all would assume that your peptide is already in the conformation which would bind, and wouldn't do any backbone flexibility.

As a final note, be careful about the initial starting location of your partners. Most Rosetta docking programs assume that the ligand and the receptor are close to each other, and frequently assume that the ligand is already in/close to the binding site. If your starting coordinates are completely separated, the docking programs may not bring them together. (e.g. if you examine your phosophocholine results, you'll see the peptide and the ligand aren't even touching. You often want to manually shift the two partners (e.g. by using editing mode in PyMol) such that they're in a "reasonable" binding conformation, or at least close.