You are here

RosettaLigand - strange score behavior

7 posts / 0 new
Last post
RosettaLigand - strange score behavior

Dear Rosetta community

I'm getting some really unrealistic scores when I'm using RosettaLigand to simulate a protein - peptide/inhibitor binding.

I'm following the guide/protocol for Rosetta 3.5 from the Documentation page along with a relax protocol. The steps I run are the following:

- python your_structure_original.pdb -ignorechain

- Using pymol to create the apo-protein and the ligand/peptide.

- Babel to convert to SDF, Avogadro to fix mistakes.


- OpeneEye - Omega with the suggested flags.

- The 'assign charges' script.

- Then the 'mol-file to params' script.


- The repack protocol (using same ex-flags as for the docking, along with suggested flags) is first run following a relaxation step (-extra_res_fa ligand.params -relax:constrain_relax_to_start_coords -relax:coord_constrain_sidechains -relax:ramp_constraints false).

- Choosing lowest energy pose with either ligand conformation 0001 or 0002, depending on purpose.

I have included the FLAGS file along with the resulting scores.
The PDB I used is 2AOD.pdb.

My interface_delta is in the millions and the fa_rep is also quite high, from what I can understand browsing the forum.

I have tried several tweekings of the FLAGS files and I've tried on many different proteins - some with inhibitors and some with native peptides (hiv-1 protease) all results with the same order of magnitude.

I'm leaving out the co-factors (waters, DMS and GOL) in the PDB - I only use the ligand (2nc) and the protease.
Am I doing something completely wrong?

Thanks for your time.

FLAGS.txt668 bytes
scores.txt176.62 KB
Post Situation: 
Mon, 2014-02-03 02:24

Typically when you get massive beneficial energies like that in binding energy, it means that you've applied a constraint that's fufilled in your bound pose, but broken in the unbound pose. The way the binding energy is calculated for ligand docking is the ligand is taken and moved ~500 angstroms away from the protein (far out of interaction distance) and the difference in energy between the two states (bound and unbound) are calculated. If there's some sort of distance constraint between the protein and the ligand, that translation can cause it to blow up in energy, making it look like there's a greatly beneficial energy of binding. If you take a close look at the if_* terms (they break down the interface delta by scoreterm) you'll see it's specifically the if_coordinate_constraint term that's giving you the large energies.

I'm not entirely sure where that term is coming from, but it may be the -tether_ligand option. Try removing that from your docking runs.

By the way, your total fa_rep isn't too high. fa_rep and fa_atr are in balance, so a higher fa_atr will almost always come along with a higher fa_rep. To see if fa_rep is high, you have to compare it to fa_atr. Yours is ~1/5 of the magnitude of fa_atr, which is entirely reasonable. Your if_fa_rep and if_fa_atr are much more closely matched, but that's somewhat expected, as you have to run thousands of ligand docks to get a reasonable number of decently scoring conformations, especially with large, flexible ligands.

Mon, 2014-02-03 15:43

Thanks a lot for your quick and informative response, Moretti.
I've started a new simulation and will return when I have some new results.

Tue, 2014-02-04 00:31

When I run without the tether_ligand option the ligand jumps out of the pocket.
I tried with some centroid start coordinates (the 'start_from' option), without tether_ligand, but it still shifted the ligand too far out of the binding pocket.
Is there a general way of choosing these centroid coordinates? I just picked a C-atom from the PDB that was 'centered' in the ligand conformation.

I'm trying to run several simulations with varying degrees of the tether_ligand option to see if I can contain the ligand within the pocket and get reasonable interface_delta values..

Thanks again for your help.

Thu, 2014-02-06 02:15

One possibility is to do the docking runs with the tether values, and then rescore the structures without the tether. If instead of "-protocol abbrev2" you use "-protocol rescore" and provide it as input the file(s) with the docking output, the ligand_dock application will skip the docking step and just rescore the structures that have already been produced.

Thu, 2014-02-06 09:41

Thanks again for your reply.
Different centroid coordinates and varying degrees of the tether_option did not solve my problem. - the more 'freedom' (the higher the tether_option value is) the better the interface_delta score is, but the more the ligand jumps out of the pocket.

I'm not sure I fully understand your suggestion, but I've extracted the structres from the first run (with low tether_option), used them as input (-s pdb1 pdb2 pdb3...) and replaced the -protocol "abbrev2" term with "rescore".

Fri, 2014-02-07 03:17

Assuming I'm doing it correctly, the numbers are getting truncated to the same value (at least in the 10 best scoring structures) and goes from -27-something million to 18-something thousands (positive), which seems disturbing.

I will try and tweek some settings but I doubt I will get anything successful.

Fri, 2014-02-07 06:48