Explicit water molecules in docking: when to keep them

Most protein structures contain hundreds of resolved water molecules, yet standard docking protocols strip them all. This blanket approach ignores a critical subset: tightly bound waters that mediate ligand–protein interactions and shape the binding landscape. Mishandling these waters leads to incorrect poses, misleading scores, and compounds that fail in the lab because a bridging water was present all along.

The role of structural waters in molecular recognition

Crystallographic water molecules fall along a spectrum. Bulk solvent is disordered and replaceable; docking software correctly models it implicitly through desolvation terms. But some waters are chemically integral to the binding site. They hydrogen-bond simultaneously to protein and ligand, acting as molecular glue. Others occupy tight pockets where ligand displacement is energetically prohibitive. Stripping these waters before docking removes real participants in the binding event.

A conserved water bridges the hinge backbone and adenine ring in many kinase–inhibitor complexes. Removing it forces the docking algorithm to propose alternate poses that ignore this energetically favorable interaction, often resulting in flipped orientations or buried polar groups. In carbonic anhydrase, a zinc-coordinated water serves as the nucleophile in the catalytic mechanism; it must remain to dock inhibitors that coordinate the metal through it. HIV protease features a pair of waters that form a hydrogen-bonded network between the catalytic aspartates and many inhibitor classes.

When waters change docking outcomes

The clearest evidence comes from re-docking experiments. Docking a ligand back into its crystallographic binding site with all waters removed often produces poses displaced by 2–3 Å or entirely incorrect orientations when a bridging water was present. Restoring that single water molecule typically recovers the experimental pose with sub-angstrom accuracy.

Scoring functions are also affected. Waters mediate polar contacts that contribute significantly to binding affinity. If you strip a conserved water, the scoring function penalizes unsatisfied hydrogen-bond donors or acceptors on the ligand, because it cannot see the water that would satisfy them. This creates artifactual rank-order changes in virtual screening: a ligand designed to use that water scores poorly, while one that ignores it may score well despite being misaligned with chemical reality.

Cross-docking into mutant structures amplifies these issues. Resistance mutations sometimes displace conserved waters by altering pocket geometry. If you carry the wild-type water coordinates into a mutant structure without checking for clashes or lost hydrogen bonds, you introduce a phantom interaction that does not exist in the mutant. Conversely, new cryptic pockets in mutants may recruit waters not present in the wild-type structure.

A practical classification scheme

Deciding which waters to retain requires inspection, not automation. We use three criteria derived from structural surveys:

• Crystallographic evidence: Waters resolved at high resolution (better than 2.5 Å) with B-factors comparable to surrounding protein atoms are more likely to be real and tightly bound. Waters with B-factors significantly higher than the local protein average are often mobile or partially occupied.
• Hydrogen-bonding geometry: A bridging water that donates or accepts at least three hydrogen bonds—at least one to protein and one to ligand or a deep pocket residue—is functionally important. Waters making fewer than two well-defined hydrogen bonds are usually displaceable.
• Conservation across structures: If the same water position appears in multiple structures of the target (co-crystal complexes with different ligands, or structures from different space groups), it is structurally conserved and should be kept.

In practice, inspect waters within 5 Å of the crystallographic ligand or the centroid of the binding site. Check PDB files for the same target with different ligands. If a water appears in the same position across three or more structures, it is almost certainly conserved. If it bridges a hydrogen-bond network and has a B-factor within 1.5× the local protein average, retain it for docking.

Implementation in Liganx Studio

Liganx Studio preserves all heteroatoms from the uploaded PDB by default, including waters. This allows manual inspection in the 3D viewer. You can toggle water visibility and selectively delete individual molecules before docking. The workflow is:

• Upload your structure and navigate to the binding site.
• Display waters as spheres and measure distances to key residues and the reference ligand.
• Check B-factors in the original PDB file (Liganx displays them in the atom property panel).
• Delete bulk solvent and surface waters far from the site; retain candidate bridging waters.
• Run docking with the retained waters treated as rigid receptor atoms.

For mutation modeling, load the wild-type structure with ligand and waters, then apply the mutation. Liganx rebuilds the mutated sidechain and relaxes surrounding residues. After relaxation, inspect whether the conserved waters still fit or clash with the new sidechain geometry. Remove any that now overlap atoms or lose their hydrogen-bonding partners. This is essential for solvent-front mutations like ROS1 G2032R or NTRK G595R, where the introduced arginine can displace a water that was present in the wild-type pocket.

Common pitfalls and how to avoid them

The most frequent mistake is treating all waters identically. Stripping everything is faster but sacrifices accuracy for convenience. Keeping everything is equally problematic: loosely bound surface waters introduce clashes and spurious hydrogen bonds. The second error is failing to validate water positions after homology modeling or mutation. A water copied from a template structure may not be stable in the modeled context.

A subtler issue arises when docking software moves or optimizes water positions during the run. Some programs allow explicit waters but treat them as flexible. If the algorithm repositions a conserved water to satisfy a ligand pose that would not occur in reality, you have replaced one approximation with another. Treat conserved waters as rigid unless you have strong evidence they adopt multiple discrete positions.

Finally, remember that docking scores are not free energies. A bridging water contributes enthalpy through hydrogen bonds but also costs entropy by restricting conformational freedom. Scoring functions approximate this balance imperfectly. A well-scoring pose with a bridging water may still fail in vitro if the entropic penalty outweighs the enthalpic gain. Use explicit-water docking to generate geometrically reasonable poses, then validate hits with more rigorous methods.

Summary

Explicit waters in docking matter when they are chemically meaningful—conserved across structures, buried in tight pockets, and integral to hydrogen-bond networks. Strip them when they represent bulk solvent or have poor crystallographic support. Inspect each binding site individually; the decision cannot be automated reliably. For wild-type docking, retain waters that bridge ligand and protein or occupy spaces the ligand cannot fill. For mutation docking, recheck water stability after sidechain rebuilding, especially for large polar substitutions. This selective approach improves pose accuracy and scoring validity without the artifacts of keeping all waters or the information loss of removing them all.

References

• Ladbury, J. E. (1996). Just add water! The effect of water on the specificity of protein-ligand binding sites and its potential application to drug design. Chemistry & Biology, 3(12), 973–980. DOI: 10.1016/S1074-5521(96)90164-7
• de Beer, S. B. A., Vermeulen, N. P. E., & Oostenbrink, C. (2010). The role of water molecules in computational drug design. Current Topics in Medicinal Chemistry, 10(1), 55–66. DOI: 10.2174/156802610790232288
• Michel, J., Tirado-Rives, J., & Jorgensen, W. L. (2009). Energetics of displacing water molecules from protein binding sites: consequences for ligand optimization. Journal of the American Chemical Society, 131(42), 15403–15411. DOI: 10.1021/ja906058w
• Ross, G. A., Morris, G. M., & Biggin, P. C. (2012). Rapid and accurate prediction and scoring of water molecules in protein binding sites. PLoS ONE, 7(3), e32036. DOI: 10.1371/journal.pone.0032036
• Barillari, C., Taylor, J., Viner, R., & Essex, J. W. (2007). Classification of water molecules in protein binding sites. Journal of the American Chemical Society, 129(9), 2577–2587. DOI: 10.1021/ja066980q