Molecular interactions

Docking, affinity and multi-molecule structure. Explore models we could test and the test families that apply to them.

Models to test

These are candidates, not a ranked list. An applicable test may need a trained head, a scoring adapter, a specialist comparison, or access to a hosted service. No result is implied by appearing here.

Native prediction
Frozen embedding + trained head
Specialist / baseline
Adapter required
Conditional access

Chai-1

released weightsfoundation model

Public code and weights under Apache 2.0; substantial compute required.

Official source ↗Paper-reported results →

Applicable tests

Biomolecular complex structure

native prediction

Native multi-component structure prediction.

Protein–ligand pose

native prediction

Joint complex output includes ligand pose.

Boltz-2

released weightsfoundation model

Public MIT code and weights; substantial compute required.

Official source ↗Paper-reported results →

Applicable tests

Protein–ligand pose

native prediction

Native complex structure prediction includes ligand pose.

Small-molecule affinity

native prediction

Native small-molecule binding-affinity output.

Biomolecular complex structure

native prediction

Native multi-component structure prediction.

DiffDock-L

2024 releasespecialist

Public pose-prediction code and weights; no native affinity prediction.

Official source ↗Paper-reported results →

Applicable tests

Protein–ligand pose

specialist / baseline

Native docking pose, not a binding-affinity value.

AlphaFold 3 Server

hosted serverfoundation model

Manual, non-commercial server access; output terms restrict automated docking combinations.

Official source ↗Paper-reported results →

Applicable tests

Biomolecular complex structure

conditional access

Manual hosted prediction; use output only within server terms.

Protein–ligand pose

conditional access

Manual complex prediction only; terms prohibit combining output with automated docking systems.

AutoDock Vina

1.2.7baseline

Public docking software; receptor and ligand preparation required.

Official source ↗Paper-reported results →

Applicable tests

Protein–ligand pose

specialist / baseline

Native docked poses; scoring-function energy is not measured affinity.

Test families

Each candidate test needs a defined dataset, split, metric and run protocol before it can become a benchmark.

Protein–ligand pose

Candidate test

Predict the bound ligand geometry from prepared molecular inputs.

Candidate models and comparators

Chai-1

native prediction

Boltz-2

native prediction

DiffDock-L

specialist / baseline

AlphaFold 3 Server

conditional access

AutoDock Vina

specialist / baseline

Small-molecule affinity

Candidate test

Predict measured binding affinity; pose confidence is not an affinity value.

Candidate models and comparators

Boltz-2

native prediction

Biomolecular complex structure

Candidate test

Predict joint structure for interacting proteins and other molecules.

Candidate models and comparators

Chai-1

native prediction

Boltz-2

native prediction

AlphaFold 3 Server

conditional access

Model sources and access descriptions checked 2026-09-16. Confirm licenses, terms and checkpoint revisions before any run.