The science behind the predictions.
Molecular learning, mechanistic interpretation and experimental feedback.
Explore the methodology
How predictions are built
Learning chemistry at molecular scale
A pretrained graph neural network learns patterns in atoms, bonds and their surroundings, turning molecular structure into a rich numerical representation.
Opsin builds on this foundation with models trained for specific biological mechanisms, connecting molecular patterns to experimental outcomes.
- Molecules in pretraining
- ~6 million
- Biological and quantum pretraining tasks
- Thousands
Molecular structure
With biological context
Learned representations
Encode chemistry and biology
Predictive models
Learn from experimental data
Mechanistic interpretation
Connect properties to response
How confidence is assessed
A prediction is only useful when you understand its limits.
Tested on unseen chemistry
Held-out chemical series challenge generalisation.
Uncertainty and applicability
See how confident a prediction is, and where the model applies.
How models improve
Each experimental round adapts the models to your programme.
Your assay results
Model adaptation
Next experiment
New results
Learn. Refine. Repeat.
Put the science to work in your next experiment.
Explore Octopus