Ligand Docking Server Documentation


RosettaLigand is a tool for docking small molecules into proteins. RosettaLigand takes as input an SDF file containing the small molecule ligand to be docked, and a PDB file containing the protein the ligands should be docked into.


  1. By default, only the provided ligand conformers will be used. If you don't want to use conformers, provide an SDF file with a single ligand conformation

  2. Conformers can be generated using OpenEye Omega, the BCL, MOE or other conformer generation tools. All conformers in the same SDF file should have the same name. You can upload a single ligand conformation and check the “Generate ligand conformers with the BCL” option to have the server automatically generate a set of conformers using the BCL.

  3. All ligand conformers in the input file must have 3D coordinates and added hydrogens. (This applies even if the server-generated conformer option is being used.)

  4. The ROSIE Ligand docking protocol is not set up to do virtual high throughput screening (vHTS). Each job submission should consist of a single small molecule being docked to a single protein. Providing multiple, chemically distinct molecules in the input SDF file will result in an error.

  5. RosettaLigand cannot perform binding site detection. The approximate location of the binding site within 5 Å should be known. SiteHound-web can can identify potential ligand binding sites with a probe molecule. The center coordinates in the SiteHound output should be entered as starting coordinates for ROSIE ligand docking. Multiple docking runs may be needed to investigate multiple potential binding sites. SiteHound-web is provided as a suggestion and is not affiliated with ROSIE in any way.

  6. While RosettaLigand is usually capable of making accurate binding predictions, some protein systems are very difficult to dock into, the following guidelines can help maximize the likelihood of obtaining high quality predictions:

    1. Docking performance his highly dependent on backbone conformation. If possible, use an X-ray crystal structure with resolution less than 2.0 Å as input

    2. Apo structures are more difficult to dock into. If possible, use an input structure co-crystallized with a bound inhibitor or native ligand. Docking performance improves if the co-crystallized ligand is similar to the ligand being docked.

    3. If multiple crystal structures of the same protein exist, dock ligands into all of them.

    4. RosettaLigand is not optimized for docking into shallow binding pockets, or predicting surface binding interactions. For best results, use relatively deep binding pockets.

    5. While RosettaLigand is capable of handling systems with co-factors, metal ions, or tightly bound waters at the protein-ligand interface, these systems are enormously more complex, and the likelihood of RosettaLigand being capable of correctly handling these systems is reduced. We strongly recommend against using this server for docking into protein-ligand systems with co-factors, metal ions, or waters at the interface.

    6. If a crystal structure with a bound inhibitor or native ligand exists, benchmark RosettaLigand by re-docking this inhibitor into the crystal structure. If the lowest scoring model is not within 2.0 Å RMSD of the crystal structure, it is unlikely that Rosetta will be capable of making accurate predictions with this protein system. See Interpreting Results for details.

    7. The RosettaLigand protocol used here typically requires about 200+ models to produce a high quality protein-ligand docking pose. (See DeLuca et al. PLoS ONE 10(7): e0132508 for performance details.)


In general, the default input parameters for the RosettaLigand server are reasonable. The parameters have the following definitions:

Advanced Settings:

The options below are considered “Advanced Settings”. Descriptions are provided along with examples of when you might adjust these settings. However, you do not need to change any of these settings in order to run ROSIE ligand docking.

Interpreting Results:

In general, the interface energy is the best metric for discriminating between ligand binding poses. RosettaLigand only minimizes protein atoms within 7 Å of the ligand while treating the rest of the receptor as rigid. However, the Rosetta energy function is evaluated as the sum of all residues in the protein, the total score is generally very noisy. Thus, we recommend that the poses with the lowest interface_delta score be selected. However, structures with abnormally high total score (as compared to the other structures in the run) may indicate a docked conformation which has contorted the protein in order to bind the ligand.

The transform_accept_ratio in the scorefile gives a rough diagnostic about how well the low-resolution stage performed. This should normally be between 0.2 and 0.8. Having more than a few structures with a transform_accept_ratio of 0 means that either the initial perturbation is set too high (for best results, this should normally be less than two thirds of the the pocket size), the grid size is too small (this should be normally set to more than the length of the ligand plus twice the pocket size), or the pocket is too small for the ligand size (try additional conformers or a different starting backbone).

If a structure co-crystallized with a bound inhibitor is present, the native ligand should be re-docked into the crystal structure using the same input settings as will be used in the experimental study. This re-docking will serve as a validation experiment to determine if RosettaLigand is capable of correctly modeling the protein system. If the lowest scoring model generated has an RMSD of more than 2.0 Å, it suggests that you are working with a protein system that Rosetta is unable to model effectively, and any predictions generated by this server should be viewed with skepticism.

When docking a ligand with unknown activity or binding position, generate at least 200 models, and select the best 1-20 models by interface_delta score (lower scores are better). The interface_delta score is the difference between the total Rosetta energy score with the ligand bound, and the ligand unbound. In general, RosettaLigand is capable of discriminating between well and poorly bound ligands based on score.

The small ensemble of top scoring models should then be evaluated visually using a tool like pymol. The predicted binding poses should be evaluated in the context of existing crystal structure information, and whatever experimental or structural data is available to you.

Alternate Protocols

RosettaLigand is intended to dock small molecule ligands only (metabolite- or drug-like organic molecules). It is not intended for docking protein, peptide, or nucleic acid ligands.

Please cite the following article when referring to results from our ROSIE server:

  1. Deluca, S., Khar, K., Meiler, J. (2015). Fully Flexible Docking of Medium Sized Ligand Libraries with RosettaLigand. PLoS ONE 10(7): e0132508. doi:10.1371/journal.pone.0132508

  2. Combs, S. A., Deluca, S. L., DeLuca, S. H., Lemmon, G. H., Nannemann, D. P., Nguyen, E. D., et al. (2013). Small-molecule ligand docking into comparative models with Rosetta. Nature Protocols, 8(7), 1277–1298. doi:10.1038/nprot.2013.074

  3. BCL conformer generation: Kothiwale, S., Mendenhall, J.L., Meiler, J. (2015) BCL::Conf: small molecule conformational sampling using a knowledge based rotamer library. J. Cheminform., 7, 47. doi:10.1186/s13321-015-0095-1

  4. Lyskov S, Chou FC, Conchúir SÓ, Der BS, Drew K, Kuroda D, Xu J, Weitzner BD, Renfrew PD, Sripakdeevong P, Borgo B, Havranek JJ, Kuhlman B, Kortemme T, Bonneau R, Gray JJ, Das R., "Serverification of Molecular Modeling Applications: The Rosetta Online Server That Includes Everyone (ROSIE)". PLoS One. 2013 May 22;8(5):e63906. doi: 10.1371/journal.pone.0063906. Print 2013. Link

We welcome scientific and technical comments on our server. For support please contact us at Rosetta Forums with any comments, questions or concerns.

Modeling tools developed by the Meiler Lab at Vanderbilt University. The Rosie implementation was developed by Samuel DeLuca, Rocco Moretti and Sergey Lyskov.