Thea unifies the full small-molecule design loop into one system. Generation, simulation, and safety modeling operate together — so candidates are optimized across every axis at once, not handed down a lossy assembly line.
The hardest targets fail conventional discovery because each step — find a hit, dock it, fix the liabilities — discards information the next step needs. Potency gets optimized, then safety breaks. Safety gets fixed, then potency collapses.
Thea closes that loop. Every candidate is evaluated against potency, selectivity, and a full safety profile simultaneously, inside a single search. The result is chemistry that is good on all axes at the same time — the only kind worth advancing.
Here is what the platform delivers — from a structurally anchored target to a cross-validated lead candidate. How it delivers it — the architectures, training strategy, and design rules — is our edge, and stays with us.
The platform is built to identify and validate disease targets and their druggable sites, then anchor design on an experimentally observed pocket — never a purely computational guess. For Titan we anchored on a target with overwhelming human-genetic validation (ApoE4) and its published co-crystal pocket.
We design novel, synthesizable molecules conditioned on a target's three-dimensional binding site — reaching original chemical matter far outside known libraries, purpose-built to fit.
Candidates are scored with structure-aware, physics-based simulation against the experimentally observed pocket — and cross-checked by independent scoring so we trust what rises to the top.
Every molecule is profiled across a full ADMET panel — cardiac (hERG), hepatic, CNS exposure, blood-brain-barrier penetrance — and screened for structural alerts and toxic motifs before it advances.
Generation, scoring, and safety feed an iterative optimizer that evolves candidate series toward the whole profile at once — converging on molecules that are potent, selective, and safe together.
The same pipeline that produced our Alzheimer's program applies to any target with an experimentally observed binding site. Steps 01–04 are complete for our flagship; step 05 is the partner-funded bridge to the lab.
Lock onto a validated target & an observed pocket.
Design novel chemistry for the pocket.
Physics-based, cross-validated ranking.
Co-optimize potency, selectivity, safety.
Partner-funded experimental validation.
We design against experimentally observed binding pockets — not computational hypotheses. If the site isn't real, the chemistry isn't either.
A molecule that fails the safety profile never advances — no matter how well it binds. Liabilities are designed out early, where it's cheap.
Top candidates are confirmed by independent scoring methods before we believe a result. Convergence across methods is the bar.
We're transparent about results and reasoning. The specifics that produce them are protected intellectual property, shared with qualified partners under NDA.
The following are deliberately not disclosed on this site:
If you have a genetically validated target with an observed pocket and a discovery problem worth solving, we should talk.
