Ligand Protonation and Tautomers for AutoDock Vina: How to Choose and Track Chemical States

This guide addresses the decision before SDF-to-PDBQT conversion: which chemical forms should enter the calculation? It is intended for researchers preparing their own small-molecule screens, particularly when one database entry can generate several docking inputs.

A source record, a modeled chemical state and a three-dimensional conformer are different objects.

Protonation changes can alter interaction patterns. Tautomers can also change which atoms act as donors or acceptors without changing the compound's net charge. The Dimorphite-DL and Gypsum-DL method papers explain why ionization and tautomer representation belong in preparation rather than being silently inherited from a database drawing. Ropp et al., Dimorphite-DL, Ropp et al., Gypsum-DL.

The practical consequence is an accounting problem as well as a chemistry problem: three prepared states do not become three independent compounds.

Write down the experiment the model is intended to support. For an assay-driven study, record the reported buffer and pH, the source of that information, and any material uncertainty. Do not silently replace missing assay information with pH 7.4.

Where the conditions are unknown, retain that uncertainty in the protocol. A provisional pH range can support a sensitivity analysis, but it should be labeled as an assumption rather than a measured property of the experiment.

Dimorphite-DL's original method uses ionizable-substructure information and pKa ranges to enumerate states for a specified pH range. This makes the model and its uncertainty relevant parts of the output, not just the pH entered into a command. A predicted state is still a model-dependent proposal. Dimorphite-DL implementation.

Use a short decision record before processing the library:

Solution-state plausibility and a possible bound state are related but not identical. A pocket may favor a representation that is less common in bulk solution. That possibility calls for a justified comparison, not unlimited enumeration or the automatic acceptance of whichever state scores best.

The AutoDock Vina tutorial places protonation checks upstream of ligand preparation and points to Molscrub for preparation of chemical variants. Its docking command does not determine a biologically correct pH-dependent state. Official Vina ligand-preparation guidance.

The tools occupy different positions:

• Molscrub documents pH-related and tautomer enumeration, as well as geometry-generation operations. Its repository describes distinct returned states that can contain multiple conformers. Record the actual version and configuration used. Molscrub official repository.
• Gypsum-DL describes a broader library-preparation workflow that generates several molecular forms and limits the resulting expansion. Its published benchmark is evidence about the tested workflow, not a promise that every additional variant improves a new screen. Gypsum-DL original method.
• Meeko parameterizes supplied molecules for AutoDock and processes docking outputs. Its preparation command is not evidence that the selected input state is biologically correct. Meeko official repository.

These are external preparation tools. Mentioning them here does not imply that MolNexus bundles their state-selection capabilities.

Formal charge belongs to the chosen chemical representation. Partial charges are assigned by a parameterization model. Changing a partial-charge field is not equivalent to selecting another protonation state.

For the standard Vina scoring function, the official FAQ states that user-supplied partial charges are ignored. That does not make protonation irrelevant: chemical preparation also affects hydrogen-bond donor and acceptor typing. Do not generalize that Vina-specific statement to every scoring function supported by the wider AutoDock ecosystem. AutoDock Vina charge FAQ.

A useful policy separates routine inputs from unresolved cases. The following is an editorial workflow recommendation, not a validated universal chemical filter.

Retain states admitted by the declared conditions and method, with correct valence, reviewed stereochemistry and interpretable provenance.

Review compounds when plausible methods disagree, relevant pKa information is uncertain, a tautomer changes a key recognition group, or the chemistry falls outside the method's documented scope. Check the underlying structure rather than voting between software outputs.

Exclude or defer malformed records, unidentified mixtures, unsupported covalent chemistry, unresolved source identity, and variants generated only to obtain a favorable score. Preserve the exclusion reason and the original compound in the denominator.

Do not automatically remove every disconnected fragment. First decide whether it is a counterion, a mixture, a required component or an incorrectly assembled record. Likewise, do not turn an explicitly defined stereoisomer into an uncontrolled stereoisomer library.

Apply the same policy to known actives, controls and prospective compounds. Special manual treatment of only the known positives can make preparation differences look like screening performance.

Do not overwrite a compound's first SDF file every time another state is generated. Keep a state manifest beside the structures.

At minimum, retain:

• Parent ID, source ID and original source-file checksum.
• State ID, state-specific isomeric representation, formal charge and stereochemistry.
• Conditions, state-generation tool, version, model and complete settings.
• Inclusion or exclusion decision, reason and unresolved uncertainty.
• Conformer ID, geometry method and random seed where applicable.
• Input SDF record, prepared PDBQT file and their checksums.
• Downstream run ID, pose ID and compound-level reporting decision.

A minimal illustration is:

These two ethylamine representations illustrate identifier and charge encoding only. They are not an assertion that both should be retained at a particular assay pH, a predicted population ratio, or a docking benchmark. A real manifest needs the conditions and justification listed above.

A second geometry of EXAMPLE001_S02 should receive a new conformer ID, not a new parent ID. A different specified stereoisomer needs an explicit identity policy; it should not be hidden as an alternative geometry.

State enumeration changes the number of opportunities each parent has to produce a favorable score. If one compound contributes one input and another contributes twenty, taking the lowest score from every parent deserves scrutiny.

Set the rules before reviewing the prospective results:

1. Define a consistent, chemically motivated state-generation policy.
2. Set an operational cap and record when it truncates the proposed set. A cap limits work; it does not establish completeness.
3. Retain state-level outputs, failures and exclusions.
4. Summarize at the parent-compound level while exposing which state drove the result.
5. Flag compounds whose position or interaction hypothesis changes substantially across retained states.

Do not average raw docking scores across states as if they were a thermodynamic ensemble, and do not interpret the score difference between two states as their solution population ratio. Such an interpretation would need a separate, justified treatment of state energetics and populations.

There is also a distinction between retrospective recovery and prospective selection. In the Gypsum-DL paper's pose comparison, the authors selected the generated variant with the lowest RMSD to the known crystal pose. That is a useful retrospective comparison, but it is not a selection rule available when the prospective binding mode is unknown. Gypsum-DL docking methodology.

A transparent shortlist can therefore say: “This parent is retained provisionally; its favorable result depends on state S02, whose relevance needs review.” That is more informative than publishing one unexplained best score.

Before conversion, inspect the state structures, verify explicit hydrogens and valid three-dimensional coordinates, and check that exported identifiers still match the manifest. After conversion, preserve both the original state SDF and prepared output. Do not use a successful parameterization step as the final chemistry review.

For an individual evaluating software, the useful operational question is whether that identity chain remains visible from input to result. MolNexus supports a local Windows docking workflow with preparation review, pose inspection, exports and history. It does not determine the biologically correct ligand state; keep the state-selection manifest alongside its outputs.

The next step is the separate conversion workflow, once the chemistry decisions are recorded:

Prepare the selected states with Meeko.