From our analysis
("Lessons learned" from Grand Challenge 2015
; updated February 2018)
From our participants
- Successful prediction of ligand-protein poses depends on the entire workflow, including factors extrinsic to the core docking algorithm, such as how the ligand and protein structures are prepared, the conformation of the protein selected, and the treatment of crystallographic waters.
- The success of docking and scoring predictions is not clearly correlated with the software used.
- Using existing structural information, such as cocrystal structures of small molecules with the target protein, can increase docking success rates. For example, known poses of similar ligands can guide positioning of the new ligand, and better docking results may be obtained by docking a new ligand into a binding site solved with another ligand with the same chemotype.
- There is a mixed picture as to whether human intervention leads to improved results.
- The accuracy of the poses used in structure-based affinity rankings does not clearly correlate with ranking accuracy.
- Explicit solvent free energy methods have not yet outperformed faster scoring methods in blinded protein-ligand affinity predictions.
(Selected responses from Grand Challenge 2
Survey, posted May 16, 2017)
- ... attending the webinar provided me with a better understanding of best practices of the different methods, as well as ideas for approaches to use going forward.
- It's given me a perspective that "more" CPU time is no always better in ligand potency prediction
- I would pay more attention to the receptor conformations and flexibility.
- It has made me more aware of the challenges of sampling. I've been working on better ways to include this into our protocols and methods.
- ... the docking poses really affect the results of binding free energy calculation so we will be more aware of that in the future
- It has made me pleasantly surprised when a scoring function actually delivers a useful result and makes me very skeptical of people who blindly trust the score that they get.
- The D3R challenges allowed us to validate our docking protocol, and the results obtained prompted us to start carrying out free energy calculations for the docking post-processing step.
- docking seems to be improved by machine learning and I plan to incorporate such approaches.
- ... it will definitely change the I do docking to avoid or minimize false positives.