What researchers are trying
IsoDDE and Boltz-2 extend co-folding systems with learned affinity heads, while hybrid studies combine AI structures with docking or physics-based free-energy methods.
Joint structure and binding-affinity reasoning
Can one model predict both the bound geometry and the energetic ranking needed to prioritize compounds?
Why it matters
Pose generation alone does not answer which molecule should be made next. Affinity, uncertainty and ranking are the decision-relevant layer.
Evidence boundary
Recent independent evaluations report that strong global correlations may not preserve ranking among top compounds, where lead-selection decisions occur.
Organizations represented
Isomorphic Labs · MIT · Open community
Demonstrated evidence
What has actually been shown.
- Developer-reported public benchmark gains for IsoDDE.
- Boltz-2 made joint open structure-and-affinity workflows broadly accessible.
Unresolved questions
- Can models rank close analogues within a chemical series?
- How well do they extrapolate across assays, protonation states and target families?
- Can uncertainty identify when physics or experiment should take over?
Signals to watch next
- Top-k enrichment
- Matched molecular-pair tests
- Prospective affinity challenges
- AI + FEP hybrid workflows
Connected evidence graph
Related BioAtlas model passports.
Boltz-1 / Boltz-2
Open-source AF3-quality structure — plus binding affinity.
4/7 evidence fields documentedAlphaFold 2 / 3
The model that solved the 50-year protein-folding problem.
4/7 evidence fields documentedSchrödinger Platform
Physics-based simulation, now fused with machine learning.
1/7 evidence fields documentedDiffDock
Reframing molecular docking as a diffusion generative problem.
4/7 evidence fields documentedPrimary and evaluation sources