Researchequivariant gnnmolecular property predictionmatlantisml potentials
Large-Scale Potentials Improve 3D GNN Molecular Prediction
6.8
Junji Seino and colleagues (submitted Feb 3, 2026) present EGNN-PFP, a framework that integrates Matlantis PFP-derived local descriptors into equivariant 3D graph neural networks to improve molecular property prediction. On QM9 the model outperforms baseline EGNNs on 11 of 12 properties, and on tmQM it improves all five target properties, highlighting enhanced modeling of transition-metal local environments.
Key Points
- 1Integrates Matlantis PFP-derived local descriptors into equivariant 3D GNNs (EGNN-PFP).
- 2Improves accuracy on QM9 for 11 of 12 targets and tmQM across all five properties.
- 3Enables better modeling of local atomic environments, notably transition-metal complexes, for predictions.
Scoring Rationale
Solid empirical improvements across QM9 and tmQM, but results are from a single arXiv preprint without peer review.
Sources
Public references used for this report.
Practice with real Real Estate data
90 SQL & Python problems · 15 industry datasets
Used by DS/ML engineers at top companies
250 free problems · No credit card
See all Real Estate problems
