Causal molecular discovery.

The Neopoly Molecule algorithm leverages experimental data and quantum-mechanical descriptors to build causal models that help you evaluate and optimize your molecule.

1. EMBED

Embed molecules into representations.

Train a hierarchical graph network within a joint-embedding predictive architecture to learn structural information; guide its learning trajectory with causal representation learning to capture causal structure-property relationships.

2. EVALUATE

Predict molecular properties.

Query the resulting causal model to predict your molecule’s property. Perform counterfactual reasoning on the causal model to assess whether your molecule is a necessary and/or sufficient cause to a certain property.

3. OPTIMIZE

Query the causal model to identify the expected or most probable representation that will yield the desired properties. Optimize your molecule to meet this optimal representation.