Evolvia AI
Binding predictions that must earn trust.
Most docking tools rank molecules by score. Evolvia AI asks a harder question first: is the binding model chemically real? It checks structures, metal geometry and controls, and it holds back any ranking until the evidence holds.
The problem
A great score can still be wrong.
Standard docking ranks poses by score. In metalloenzymes, the chemistry around the metal decides whether a pose is real. Flip between the two approaches using our own HDAC6 benchmark.
HDAC6, 5EDU/TSN, five seeds
A score-only pipeline would pass this. The wrong zinc geometry goes unnoticed.
Evolvia AI measures the zinc geometry separately. The result blocks ranking and is logged as a method limit.
Illustration of the decision logic using our benchmark counts. It is a method result, not a statement about any compound's potency.
Metal-aware
Eight metals. Eight sets of rules.
A zinc site and a magnesium site do not coordinate alike. Evolvia AI uses the right donor rules for the metal in front of it, and checks the geometry against the structure.
- Donor count and typeAre the right atoms bound, and enough of them?
- Donor-to-metal distancesMeasured from the coordinates, not assumed.
- Coordination geometryDoes the arrangement fit the metal?
- CompetitionWater and ligand atoms that crowd the site.
Typical values from general bioinorganic chemistry. They act as sanity checks against the experimental structure and do not override it.
Docking engines
One router. Three engines. One verdict.
Evolvia AI works above AutoDock Vina, GNINA and AutoDock4Zn. It sends each compound to the engine that suits the target chemistry, then compares what comes back.
Live validation ledger
Evidence before claims.
Status reflects completed computational gates, not biological efficacy or patentability. Failures and limits are listed as plainly as passes.
HDAC structural qualification
Human HDAC6-CD2 (5EDU), human HDAC8 (1T69), and an HDAC6-CD1 domain comparator passed catalytic-zinc and crystallographic-ligand checks.
Redocking with crystal contacts
On a second target, the native ligand was redocked in five of five seeds within 2.0 Å after crystal-symmetry neighbours were added to the receptor. This is a method check, not evidence of potency.
Vina pose recovery
For 5EDU/TSN, five of five seeds recovered the native orientation within 2.0 Å, while zero of five reproduced the required bidentate Zn geometry.
Metal-aware confirmation
AutoDock4Zn confirmation and HDAC8 receptor-provenance compatibility must be completed before enrichment or prospective ranking.
Candidate ranking
Ranking remains blocked pending pose recovery, matched active-decoy enrichment, isoform counter-screens, biochemical calibration, and IP review.
What you get
Results you can audit.
Every campaign leaves a record that a scientist, a partner or an investor can check.
Receptor provenance
Which structures were used, what was repaired, and why.
A full decision log
Every decision, pass or fail, with the evidence behind it.
Control results
Redocking and enrichment controls reported before any candidate.
Assay-ready hypotheses
A shortlist for the bench, only when the gates allow one.
Work with us
Bring a difficult target.
Evolvia is built for traceable, experimentally grounded discovery, especially where metal chemistry or incomplete structures make conventional scoring unreliable.
Scope the target
We review the target, the structures and the question you need answered.
Validate before ranking
Controls run first. Only then do candidates move forward.
Hand off to the bench
You receive testable hypotheses, the gate log and the open questions.
Questions, answered plainly.
Does Evolvia AI predict potency?
No. It prioritizes testable hypotheses and checks whether a binding model is chemically credible. Potency and selectivity are decided by enzyme assays, and the platform is built to learn from those results.
What happens when a gate fails?
The campaign stops advancing and the failure is logged. A favorable docking score cannot override a failed chemistry check. That is what "fail closed" means.
Which targets is it built for?
Targets where metal chemistry or incomplete structures make conventional scoring unreliable, such as zinc and iron enzymes. It is target-agnostic by design, and each new target starts with its own qualification and controls.
Is Evolvia AI available as software?
Access today is through collaboration, after a technical review of the target and the data. The three routes are co-discovery, platform access, and our own pipeline.
How do I know the results can be trusted?
Every receptor state, repair, control and decision is recorded, including the failures. The validation ledger on this page lists what passed, what limited us, and what is still locked.
Have a target where scores keep misleading you?
Tell us about it. We will tell you what the platform can and cannot answer.