How to Select Compounds After Molecular Docking: From Ranked Results to an Experimental Shortlist

A study published online in July 2026 modeled three large-scale docking campaigns in which 2,682 molecules had been synthesized and tested across poor, intermediate, and favorable scoring regions. The authors treated docking score as a noisy correlate of binding affinity and added a term for high-ranking artifacts. Their model reproduced the observed hit-rate curves and predicted that artifacts can increasingly dominate the most extreme scores as libraries grow [1].

That result does not make docking rank useless. In a separate prospective study, Liu and colleagues docked 1.7 billion molecules against AmpC beta-lactamase and tested 1,521. In that campaign, hit rates and affinities improved with docking score, while estimates of hit rate stabilized only after several hundred molecules had been tested [2]. Read together, the studies support a practical position: ranking can contain useful campaign-level signal, but the first rows alone are not a complete experimental-selection strategy.

A shortlist for twelve purchasable compounds is a different decision from a 384-well primary assay or a synthesis campaign. State what the experiment must learn: detect binding, test a functional response, compare chemotypes, challenge a predicted interaction, or establish whether a screening protocol deserves a larger round.

Then freeze the practical boundary. Record the number of assay slots, the target construct and assay format, compound-source requirements, controls, test concentration or dose-response plan, confirmation method, and the result that would justify the next step. There is no universal number of compounds that makes a docking shortlist valid. The right number depends on the decision, assay variability, available controls, and resources.

Start selection from a traceable screen, not a detached CSV of successful scores. Keep every submitted ligand identity, preparation outcome, docking failure, receptor record, search box, engine version, scoring function, seed policy, pose, and export connected to the job. Otherwise, filtering can silently change the denominator and make a later shortlist impossible to reconstruct.

The dedicated guide to traceable AutoDock Vina batch docking explains this execution record. If the protocol itself has not yet been challenged, use the pre-screening validation workflow before treating its ranking as selection evidence.

A selected row must map back to the exact chemical entity that will be ordered, synthesized, or retrieved. Confirm source ID, structure, stereochemistry, protonation and tautomer policy, salt or fragment handling, prepared 3D state, and PDBQT provenance. Do not let a convenient supplier record silently replace the state that was docked.

Check whether the compound can actually enter the experiment: supplier or synthesis status, expected lead time, quantity, purity information, solvent compatibility, and any known stability or solubility constraint. These checks do not predict biological success; they prevent the computational candidate and tested material from becoming different objects. For file-level problems, use the SDF preflight checklist.

High-ranking candidates deserve attention because a validated campaign may enrich useful chemistry near the top. They should not receive automatic assay priority merely because their numerical order is visible.

Moesgaard and colleagues represented a molecule's position by a log-normalized rank and argued that physically testing compounds across a range of ranks is needed to locate the peak hit rate for a specific campaign [1]. This is not a new AutoDock Vina parameter or a universal instruction to test weak poses. It is an experimental design principle: when resources permit, deliberate rank-band sampling can reveal whether the extreme top is enriched, flat, or contaminated by artifacts. Record the rank-selection rule before assay results are available.

Review more than the best numerical mode. Examine severe clashes, buried unsatisfied charges, improbable ligand geometry, exposure of hydrophobic groups, receptor artifacts, and whether the pose supports interactions justified by structural, mutational, or ligand evidence. A generic hydrogen-bond count is not a substitute for a target-specific binding hypothesis.

Prospective studies commonly add this layer. Grotsch and colleagues nominated compounds for synthesis using docking score, predicted pose, chemical novelty, and diversity [3]. Zhou and colleagues filtered and clustered candidates, manually reviewed favorable interactions and geometries, then synthesized and experimentally tested the selected set [4]. These examples support multi-criterion selection in their reported targets; they do not define a universal scoring formula.

Use the separate guide to interpreting AutoDock Vina affinity, RMSD, and poses when the result review is not yet complete.

If every assay slot is occupied by close analogs, one chemotype can dominate the answer. Define how similarity will be measured before selecting representatives: fingerprint, similarity coefficient, scaffold definition, clustering algorithm, and any threshold. Bemis and Murcko introduced a framework-based way to organize ring and linker cores that remains one useful view of structural diversity [8]. Fingerprint and scaffold views answer different questions, so retain the chosen method rather than reporting that the set was merely “diverse.”

Selecting representatives from distinct clusters broadens the structural hypotheses tested. A small number of close analogs can still be valuable when the explicit goal is early structure-activity information. Keep discovery diversity and series expansion as separate objectives instead of blending them after the results are known.

Structural alerts can identify candidates that deserve additional scrutiny. The original PAINS filters were derived from frequent hitters observed in one assay-detection technology, although the authors discussed the patterns across other assays [5]. An alert is therefore a reason to inspect the compound and assay context—not a stand-alone experimental conclusion.

Colloidal aggregation is another source of artifactual inhibition. Irwin and colleagues assembled more than 12,000 known aggregators and built a precedent-based advisor, while explicitly warning that the goal was not to eliminate every potentially aggregating chemotype [6]. Use predictions and similarity warnings to plan appropriate controls with the assay team. A clean result requires experimental behavior, not the absence of a software alert.

A candidate list and an assay plan should be designed together. Include the relevant reference ligand or positive control, suitable negative and technical controls, repeat or dose-response criteria, and the orthogonal measurement required before a signal advances. In the RosettaVS study, initial competition-assay findings for a KLHDC2 candidate were confirmed with biolayer interferometry, and a crystal structure later tested the predicted pose [4]. The sequence of evidence—not one docking score—made the result informative.

Docking and experimental screening can also contribute different chemotypes. In a parallel cruzain campaign, Ferreira and colleagues found partly distinct series from docking and high-throughput screening, while joint evidence helped prioritize well-behaved candidates [7]. The practical implication is not that every project needs both technologies. It is that orthogonal evidence can expose what one method misses.

MolNexus 0.1.1 is BioChemIntelli's local Windows desktop application for the protein-small-molecule docking layer. It keeps receptor and ligand preparation review, interaction-box definition, AutoDock Vina execution with Vina or Vinardo scoring, pose inspection, local job history, and exports in one workspace. That supports the upstream evidence record needed before an experimental shortlist is finalized.

MolNexus does not automatically choose experimental candidates, cluster chemical diversity, predict assay interference or ADME, procure compounds, or run an assay. Those decisions remain with the researcher and any chemistry or experimental team. Workflow organization improves visibility and repeatability; it does not make a docking score more accurate.

The current commercial profile is a US$499 one-time license for one Windows PC at a time, perpetual use of the purchased version, and 12 months of updates. MolNexus is coming soon; purchase and download are not open yet.

The best post-docking shortlist is not the table with the most negative scores. It is a prespecified, traceable allocation of experimental capacity across rank signal, pose plausibility, chemical diversity, artifact risk, availability, and controls.

Preserve every denominator, give each selected compound a scientific purpose, and connect measured outcomes back to the exact computational record. That turns a docking screen into a learning cycle: not proof before the assay, but a stronger way to decide what the assay should test next.