The challenge with AI-driven material discovery
Generative models for open-ended materials exploration (Figure 1) suffer from poor extrapolation of OOD material properties (Figure 2). A key challenge is to identify their applicability domain and reduce this OOD generalization risk while providing prediction uncertainties.
Neopoly leverages causal representation learning with small datasets to build interpretable causal models that uncover underlying relations between molecular representations, quantum-mechanical descriptors, and material properties.
Advancing material discovery AI with causality
Use Neopoly Molecule to evaluate and optimize your molecules, and Neopoly Formulation to find your optimal chemical formulation.

Neopoly on AWS

Figure 1: Baseline QM-assisted GNN models used to predict molecular properties

Figure 2: Extrapolation of property range for Tb, comparing small and large data sizes for training.

Shimakawa, H., Kumada, A. & Sato, M. Extrapolative prediction of small-data molecular property using quantum mechanics-assisted machine learning. npj Comput Mater 10, 11 (2024). https://doi.org/10.1038/s41524-023-01194-2

The challenge with AI-driven material discovery

Generative models for open-ended materials exploration (Figure 1) suffer from poor extrapolation of OOD material properties (Figure 2). A key challenge is to identify their applicability domain and reduce this OOD generalization risk while providing prediction uncertainties.

Neopoly leverages causal representation learning with small datasets to build interpretable causal models that uncover underlying relations between molecular representations, quantum-mechanical descriptors, and material properties.

Advancing material discovery AI with causality

Use Neopoly Molecule to evaluate and optimize your molecules, and Neopoly Formulation to find your optimal chemical formulation.

hello@neopolyai.com