I was wondering whether there is any limitation of the number of conformers for Rosetta Ligand docking. I have a large ligand with 25 rotatable bonds and generated top 500 conformers with BCL. The reason I picked 500 conformers is that I saw in Rosetta papers they worked with 100 conformers for a small molecule with around 5 rotatable bonds, so I thought I need to increase the number of conformers propothional to the number of rotatable bonds? Is this correct or do I need to include all the conformers? Moreover, should I change any parameters in Transform mover like the number of cycles due to the large conformers I have?
One more question: I read a paper stating that one should split the large ligand into fragments before parameter generation. I tried to do that but molefile_to_params.py failed to generate parameters. I am not sure whether that statement is still valid.
What extra steps I should follow when I work with large ligands in Rosetta ?
Any suggestion would be highly appreciated.
There's not really a theoretical limit on the number of rotamers, it's just that with enough rotamers the ligand docking process slows down a bunch, and the sampling becomes an issue.
Usually ~200 (heavy-atom) conformers is a good result. It depends a bit on the conformational generation program you use, but usually 200 conformers is enough such that at least one of them is reasonably close to the native conformer (https://pubmed.ncbi.nlm.nih.gov/33351632/)
I mentioned heavy-atom conformers because there's additional rotamer elaboration with rotatable hydrogens (mostly hydroxyl hydrogens). There's some limits internally in Rosetta to avoid bloating the set of rotamers too much with such hydrogen rotamers. (Which can happen with hydroxyl-heavy residues.) If you have too many heavy-atom rotamers, you may see reduced amounts of hydroxyl rotamer sampling because of this.
At one point in time, there was some (rudimentary) support for splitting large ligands into multiple smaller ligands. I haven't ever attempted this, and am not really aware of people who have. One issue here is that the movers which sample the ligand rigid body positions have to be adjusted to move multiple residues, and I'm not sure how robust all of them are. The other issue is that there are techincal limitations with sampling the rotatable bonds which bridge the atoms between the connection points. (In the same way that protein backbone sampling is treated differently than sidechain sampling.) -- You can certainly try it, but it would probably be a bunch of troubleshooting and protocol (re)development to make things work.