MolVerity AI’s first implemented module: reliability-centered computational toxicity screening for early small-molecule discovery, developed within a research framework that evaluates classical machine-learning QSAR models, fingerprint-based neural networks and graph neural network architectures.
ToxVerity currently demonstrates the MolVerity reliability framework across Ames mutagenicity, hERG blockade, SR-ARE, SR-ATAD5, SR-MMP and SR-p53. These six endpoints are the current benchmark panel, not the permanent endpoint limit of ToxVerity.
Bacterial mutagenicity-related activity for early genotoxicity screening.
Potential cardiac hERG potassium-channel blockade.
Oxidative-stress response through the antioxidant response element.
DNA-damage-associated stress response.
Mitochondrial membrane-potential disruption.
p53 stress-response pathway activation.
ToxVerity is not conceptually limited to six endpoints. As the model library expands, additional validated toxicity endpoints can be incorporated into configurable customer assessment panels under the same reliability-centered workflow.
Additional endpoint models can be introduced when appropriate datasets, validation evidence, calibration, applicability-domain assessment and decision criteria are established.
Different programs can select different available validated endpoints according to their scientific and development priorities.
ToxVerity is designed to report not only what the model predicts, but how much support exists for that prediction.
Probability after calibration.
Endpoint-specific decision threshold.
Ensemble disagreement.
Representation by development chemistry.
Similarity to training chemistry.
Novel structural scaffold detection.
Distribution-shift warning.
No supported binary call when evidence is insufficient.
SMILES or molecular library.
Reproducible structure processing.
Endpoint-specific inference.
Probability interpretation.
Domain, novelty and uncertainty.
Experimental follow-up recommendation.
The production customer platform is not yet publicly open. Access is currently arranged for research pilots and evaluation engagements.
