Independently validated.
Competitively proven.
Anyone can claim their model is accurate. Blind prediction competitions run by independent scientific bodies are a different standard entirely — one we've met twice.
Why blind competitions matter more than internal benchmarks
Self-reported model accuracy is easy to manipulate — intentionally or not. Competitions run by independent scientific organisations change that dynamic.
Blind test compounds
Competitors do not know the test compounds in advance. Predictions are submitted before results are revealed — there is no opportunity to cherry-pick.
Independent judges
Results are evaluated by independent scientific bodies with no commercial interest in the outcome. This is external, third-party validation.
Global competition
Competing against the best in silico teams worldwide — academic groups, commercial tool providers, and research institutions.
2nd Place — International AI Toxicology Competition
The Tox24 Challenge is an international blind prediction competition designed to assess progress in computational methods for predicting in vitro activity of compounds — specifically, interference in biochemical pathways using only chemical structure data. Organised by the AIDD network, AiChemist, and Chemical Research in Toxicology, it crowdsources independent approaches to predict Transthyretin (TTR) binding activity across the EPA's Tox21 compound library.
Out of all participating international teams, Toxometris ranked 2nd place overall — an external, independent evaluation of our model's predictive accuracy on compounds never seen before submission.
Blind prediction of toxicity endpoints for unseen chemical compounds, evaluated independently against experimental results.
European prediction challenge with independent blind-test validation across multiple toxicity and ADMET endpoints.
Group Winner — 2nd EUOS/SLAS Joint Challenge
The EUOS25 Challenge, co-organised by EU-OPENSCREEN, SLAS, and the AiChemist network, assessed computational methods for predicting absorption and fluorescence properties of compounds across four subtasks — using a dataset of approximately 100,000 compounds systematically tested under the EU-OPENSCREEN bioprofiling programme. Predictions had to be derived from chemical structure alone, on a blind test set withheld until after submission.
Toxometris won its group in the EUOS25 Challenge — a second independent external validation of our models, on a distinct set of endpoints from Tox24.
What this means for your team
When you use Toxometris, you're not relying on self-reported model accuracy. Our performance has been tested against real compounds, by independent scientists, in competition with the best in silico teams globally.