Cormorant: Covariant Molecular Neural Networks
By Brandon M. Anderson, Truong Son Hy, Risi Kondor. 2019
498 citations recorded by the source corpus.
We propose Cormorant, a rotationally covariant neural network architecture for learning the behavior and properties of complex many-body physical systems. We apply these networks to molecular systems with two goals: learning atomic potential energy surfaces for use in Molecular Dynamics simulations, and learning ground state properties of molecules calculated by Density Functional Theory. Some of the key features of our network are that (a) each neuron explicitly corresponds to a subset of atoms; (b) the activation of each neuron is covariant to rotations, ensuring that overall the network is fully rotationally invariant. Furthermore, the non-linearity in our network is based upon tensor products and the Clebsch-Gordan decomposition, allowing the network to operate entirely in Fourier space. Cormorant significantly outperforms competing algorithms in learning molecular Potential Energy Surfaces from conformational geometries in the MD-17 dataset, and is competitive with other methods at learning geometric, energetic, electronic, and thermodynamic properties of molecules on the GDB-9 dataset.
Subfields: Computational Theory and Mathematics, Materials Chemistry, Molecular Biology, Neural Architecture Search & AutoML, Computational Chemistry & Drug Discovery
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Computational Theory and Mathematics
- Tess E. Smidt
- Gisbert Schneider
- Michael Gastegger
- Francesca Grisoni
- K. Müller
- Erik J. Bekkers
- Kristof T. Schütt
- Mario Geiger
In Materials Chemistry
- Tess E. Smidt
- K. Müller
- Mario Geiger
- Michael Gastegger
- Max Welling
- Kristof T. Schütt
- Oliver T. Unke
- Yi Liao
In Molecular Biology
- Tess E. Smidt
- Max Welling
- K. Müller
- Michael Gastegger
- Stephan Günnemann
- Gisbert Schneider
- Francesca Grisoni
- Oliver T. Unke
In Neural Architecture Search & AutoML
- Maurice Weiler
- Max Welling
- Ron O. Dror
- Tess E. Smidt
- Stephan Eismann
- Raphael J. L. Townshend
- Mario Geiger
- Gabriele Cesa
In Computational Chemistry & Drug Discovery
- Tess E. Smidt
- K. Müller
- Mario Geiger
- Stephan Günnemann
- Johannes Klicpera
- Kristof T. Schütt
- Michael Gastegger
- Oliver T. Unke
Citations in the corpus
- E(n) Equivariant Graph Neural Networks (2021, methodology): “For QM9 (Ramakrishnan et al., 2014) we used the dataset partitions from (Anderson et al., 2019).”
- Directional Message Passing for Molecular Graphs (2020, methodology): “, 2018) or by using a fully Fourier space model (Kondor et al., 2018; Anderson et al., 2019).”
- Relevance of Rotationally Equivariant Convolutions for Predicting Molecular Properties (2020, background): “Support for Cormorant’s so-called two-body interaction has also been included in e3nn but is not considered in this experiment.”
- SE(3)-Transformers: 3D Roto-Translation Equivariant Attention Networks (2020, neutral): “[1] for 6 regression tasks in Table 2.”
- Generalizing Convolutional Neural Networks for Equivariance to Lie Groups on Arbitrary Continuous Data (2020, methodology): “We use the same train, validation, test split as Anderson et al. (2019), with 100k molecules for train, 10% for test and the remaining for validation.”
- General E(2)-Equivariant Steerable CNNs (2019, methodology): “The strategy followed in [39, 33, 20, 34] is based on well known analytical solutions and does not generalize to arbitrary representations.”
- Learning from Protein Structure with Geometric Vector Perceptrons (2020, methodology): “The equivariance of GVP layers with respect to 3D translations and rotations also highlights a similarity to methods that leverage irreducible representations of SO (3) to define equivariant convo-lutions on point clouds (Thomas et al., 2018; Anderson et al., 2019).”
- E(3)-equivariant graph neural networks for data-efficient and accurate interatomic potentials (2021, methodology): “works [11, 18, 19] demonstrates capability to perform molecular dynamics simulations.”
- Geometric Prediction: Moving Beyond Scalars (2020, background): “There has recently been significant excitement around equivariant networks that have the ability to reason about arbitrary tensors [3], [16], [4], [5], [17], [18], and though they do not predict tensors, such methods have demonstrated significant results on scalar prediction problems in medical imag”
- Integrating Scientific Knowledge with Machine Learning for Engineering and Environmental Systems (2020, background): “1) [266] [10] [219] [158] [140] [182] [203] [76] [86] [130] [174] [204] [231] [313] [121] [125] [187] [92] [306] [180] [298] [184] [267] [39] [247] [133] [130] [231] [298] [173] [66] [24] [255] [12] [17] [203] [204] [213] [239] [125] [281] [214] [176] [249] [310] [309] [311] [245] [78] [266] [297] [”
- The design space of E(3)-equivariant atom-centred interatomic potentials (2022, unknown): “We denote the set of neighbors of an atom i , i.e., atoms with the cutoff distance, by N ( i ).”
- Hierarchical, rotation‐equivariant neural networks to select structural models of protein complexes (2020, background): “32 33-36 Specifically, tensor field networks 34 have two key properties that distinguish them from conventional CNNs: they operate on a set of points with coordinates in 3D space (not a discretized grid), and they are automatically equivariant to rotations, translations, and per-mutations of those…”