AI for Bio · Model

Vina

Classical docking program, not deep learning, and the field's de facto baseline

Structure & co-foldingAcademic labUnited StatesMaintained only中文

Overview

NameVina · aliases: AutoDock Vina, Vina 1.2.x
Full nameAutoDock Vina
Ambiguous names (not auto-linked)Vina, AutoDock
FamilyNot a family entry. A single program; the other components of the AutoDock Suite are outside this record
OrganizationScripps Research
Organization typeAcademic lab
CountryUnited States
Model classClassical docking and scoring function
DomainStructure & co-folding (also: Small molecules)
First release · journal
Basis: Online date of the original paper (the issue is dated 2010-01-30)
Note: The site said 2010, which is the issue year source
Other milestones
  • First public evidence: · Original paper online source
  • Journal online: · Original paper online source
  • Code release: † · The year of the 1.2.0 paper; the repository release date is unverified source
Latest update† partly verified · code release
Basis: GitHub release v1.2.7
v1.2.7 was published 2025-02-26 after v1.2.6 on 2025-01-13; commits through 2026-10-15 are maintenance
Note: The release notes were not read, so the description is partly verified source
The most recent sourced event, not a claim that nothing changed since.
StatusMaintained only · Tagged releases, with recent commits as maintenance
OpennessCode: Open · Apache-2.0 (per the repository)
Weights: Open
Source is open; there are no deep-learning weights, and the scoring-function parameters ship with the source
Benchmarks reported2 counted (retrospective 2, cited only 1)
Wet-lab validationNone recorded
Unverified fields† release_milestones.code_release.date, latest_update.description
As of2026-11-16

What it does

Finds ligand poses inside a given pocket box with an empirical scoring function and iterative local search; it is fast and easy to run. In AI comparisons it is the floor baseline for physics scoring with a known pocket, set against the generative docking of DiffDock and against co-folding models.

Benchmarks reported

Every row is one pairing from the survey's relation file; a score belongs to the benchmark page's ranking, where it is shown with metric, eval set, source type and date. 2 of 3 pairings are counted.

BenchmarkRelationEval set / protocolSource typeNote and evidence
12 DUD-E / LIT-PCBARetrospective dataset useLIT-PCBA, rerun by a third-party studyThird-partyThird-party LIT-PCBA study ran AutoDock/GNINA †partly verified · evidence
33 PoseXRetrospective dataset usePoseX subset, evaluated by the benchmark authorsThird-partyEvaluated by the benchmark paper on its Astex subset. †partly verified · evidence
05 PoseBustersCited as comparison only (not counted)PoseBusters V2 with known pocket, from a third-party comparison tableThird-partyQuoted from the comparison table of another model's paper †partly verified · evidence

Other evaluations (outside the 44)

  • Role as the traditional-method baseline (third party evaluation) Third-party: It recurs as the floor baseline in deep-learning docking and co-folding papers, usually with the pocket known
  • How the scoring function was fitted (self report) Self-reported: The original paper states the scoring function was fitted on a public refined complex set using more than linear regression (authors' account)
  • Suite maintenance (resource use) Self-reported: The AutoDock Suite is developed jointly by four laboratories and coordinated by the team at this record's institute

Developers report: Developers report fast pose generation inside a given pocket from an empirical scoring function plus iterative local search, with a low barrier to use (per the original and 1.2.x papers).

Independent evidence: Third-party papers and benchmark authors have rerun it on several datasets, with conclusions that shift by task and pocket setting (no figures).

Training data

Group in Figure 7: Public data only.

ClassValueDetail
a. Public databasesYesThe scoring function was fitted on the PDBbind refined set; there is no large deep-learning training set source
b. Private or licensed dataNo
c. Own wet-lab data used for trainingNo
d. Wet lab used only for validationNo

Wet-lab validation

None. The developers report no wet-lab validation.

Background

  • Organization: An academic project maintained as open source by the institute's docking-tool team in La Jolla, California; the AutoDock Suite is developed jointly by four laboratories under that team's coordination.
  • Funding: Academic project; funding amounts undisclosed
  • Products and partnerships: Version 1.2.0 was published as a methods paper in 2021 and minor releases continue; it is the most common traditional comparator in docking and co-folding papers.
  • Related models: DiffDock, NeuralPLexer3, AlphaFold 3

Reading notes

  1. Its figures assume a known pocket while co-folding models mostly predict blind; the conditions differ.
  2. Classical scoring is not penalised on physical validity; in virtual screening it depends on the task.
  3. It works as a floor baseline, not as a primary competitor.
  4. It is not deep learning, so there are no weights or training scales to compare.

Caveats

  1. The developers never self-report on the site's benchmarks; every relation comes from third parties.
  2. The latest release's notes were not read, so the description is only partly verified.
  3. The name is a common word and must pass the same-name guard before auto-linking.

Sources

  1. Original docking paper (journal) ·
  2. Methods paper for version 1.2.0 (journal)
  3. AutoDock Vina GitHub releases (repository)
  4. Official AutoDock Suite page (official site)
  5. Third-party comparison table in a co-folding paper (third-party paper)

Corrections

3 corrections
FieldWasNowBasisFound
first_release2010 (the journal issue year)2009-06-04, the original paper's online datesource2026-11-16
latest_update2025-03-092025-02-26, the publication date of v1.2.7source2026-11-16
latest_update.descriptionthe previous release v1.2.6 was 2025-02-15the previous release v1.2.6 was 2025-01-13source2026-11-16

Part of the AI for Bio benchmarks survey · all 112 models · open in the explorer. Record as of 2026-11-16.