Equivariant Graph Networks for Molecules
Graph neural networks that respect rotations and translations of atoms, used to predict molecular properties and dynamics.
Tier Grades
Tier A: Frontier AI labs
feasible
Microsoft Research and universities publish these architectures.
Tier B: Indonesian AI labs
feasible
Small architectures on public benchmarks fit a modest budget.
Tier C: Institution units such as ITB
feasible
Small architectures on public benchmarks fit a university chemistry or physics lab.
Tier D: Organisations such as SEMAI (PoT, PoC, PoV)
stretch
Property prediction for a formulation is possible but needs domain data.
Where this comes from
Tiers: a) frontier AI labs, b) Indonesian AI labs, c) institution units such as ITB, d) organisations such as SEMAI delivering proof of technology, concept and value. Each rationale cites the anchors below.
Anchors
- ComputeEGNN avoids costly higher-order representations; compute not stated in the abstract.
- DataMolecular property benchmarks; SchNet models energies and forces of molecules.
- Team size (proxy)Proxy only (author count of an exemplar paper, not a headcount): 3 authors on the EGNN paper (arXiv 2102.09844); 6 on SchNet.
- Time horizonYears.
- Shelf-life signalsThese architectures became the building blocks of universal interatomic potentials on Matbench Discovery.
- Pipeline roleUniversities publish architectures; Tier B and Tier C train them on local chemistry; Tier D could use them in formulation proofs.
Where this comes from
Team size is a proxy (author counts of exemplar papers), not a headcount.
Who and What
Exemplars
Observed actors
University of Amsterdam
Technische Universitat Berlin
Technical University of Munich
Microsoft Research
University of OxfordLinked models
Links
R&D TopicEquivariant Graph Networks for Molecules
ModelsEGNN (E(n) Equivariant Graph Neural Network)University of Amsterdam, scientificSchNetTechnische Universitat Berlin, scientificDimeNet and DimeNet++Technical University of Munich, scientificViSNetMicrosoft Research, scientific+1 more
EGNN (E(n) Equivariant Graph Neural Network)University of Amsterdam, scientificSchNetTechnische Universitat Berlin, scientificDimeNet and DimeNet++Technical University of Munich, scientificViSNetMicrosoft Research, scientific+1 moreUse Casesnone recorded
Who Runs Itnone recorded
Models 5
SE(3)-TransformerUniversity of Oxford, Bosch, scientificproduced byDimeNet and DimeNet++Technical University of Munich, scientificproduced byEGNN (E(n) Equivariant Graph Neural Network)University of Amsterdam, scientificexemplarSchNetTechnische Universitat Berlin, scientificexemplarViSNetMicrosoft Research, scientificproduced byPapers
- Tensor field networks: Rotation- and translation-equivariant neural networks for 3D point cloudsNathaniel Thomas et al., arXiv preprint, 2018 founding
- Neural Message Passing for Quantum ChemistryJustin Gilmer et al., arXiv preprint, 2017 founding
- MACE: Higher Order Equivariant Message Passing Neural Networks for Fast and Accurate Force FieldsIlyes Batatia et al., arXiv preprint, 2022 state of the art
- E(n) Equivariant Graph Neural NetworksVictor Garcia Satorras et al., arXiv preprint, 2021 state of the art
- Geometric Deep Learning: Grids, Groups, Graphs, Geodesics, and GaugesMichael M. Bronstein et al., arXiv preprint, 2021 survey
Where this comes from
The papers to read first: where the direction began, its current best, and a survey when one exists.
Sources
- arxiv.org/abs/2102.09844 https://arxiv.org/abs/2102.09844
- arxiv.org/abs/1706.08566 https://arxiv.org/abs/1706.08566
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Equivariant Graph Networks for Molecules", https://aurion.semai.id/topics/equivariant-graph-networks-for-molecules/, records of October 2026.