QuantamathicalShow the result.
Keep its limits in view.
Different results answer different questions. Explore the evidence below.
Computed profiles, with measured endpoint labels where available.
Connectivity overlaps excluded from the evaluated set.
Ranking performance in that evaluation, not percentage accuracy.
What the hERG evaluation does—and does not—show.
The saved evaluation concerns an ADMET-AI-based ranking workflow. Its endpoint is related to the model endpoint, and the evaluation is not scaffold-held-out. It does not establish clinical safety, broad out-of-domain reliability, or a wholly independent prospective test.
A docking score needs context.
No reviewed target passport had full GO status. A favorable score should not outrun the method's evidence.
| Target | Reviewed status | Interpretation |
|---|---|---|
| CDK2 | Caution | Promising retrospective ranking; pose validation remains incomplete. |
| FabI | Not qualified | Insufficient separation from simple ranking baselines. |
| LRRK2 | Not qualified | Geometry checks did not translate to useful ranking. |
| InhA / GSK3B | Not qualified | Redocking checks did not meet the reviewed criteria. |
A shortlist is
a hypothesis to test.
The reviewed molecular series and internal shortlists are computational proposals. Their synthesis and biological activity are not established by the reviewed experimental evidence.
How should Atlas agreement be read?
The historical roughly 89% label agreement describes in-domain reproduction across available benchmark labels. It is not independent proof of performance on new chemistry. Measured labels exist for only a subset of records and endpoints.
What does a novelty search establish?
Whether a match was found in the sources searched, under the comparison used. A negative search is not proof of global novelty or freedom to operate.
What remains experimental?
Synthesis, biochemical activity, selectivity, broader biological behavior, and safety all require appropriate experimental evidence.
Published tools.
Connected research workflows.
Quantamathical develops workflow integration and research-specific checks around tools with their own authorship and licenses.
| Method | Role in the broader stack | Source |
|---|---|---|
| RDKit | Molecular descriptors and cheminformatics | RDKit |
| ADMET-AI | Predicted molecular properties and endpoints | Swanson et al. |
| REINVENT4 | Separate upstream AI generation workflows | MolecularAI |
| AutoDock Vina / Vinardo | Docking and scoring | AutoDock Vina |
| AiZynthFinder | Retrosynthesis proposals | MolecularAI |
| Myokit | Electrophysiology simulation infrastructure | Myokit |
| PySCF / Qiskit | Classical and quantum chemistry research | PySCF · Qiskit |
Have a question about
the evidence?
Discuss the methods, evaluation scope, or a research collaboration.