GeoMol: Torsional Geometric Generation of Molecular 3D Conformer Ensembles

By Octavian-Eugen Ganea, Lagnajit Pattanaik, Connor W. Coley, Regina Barzilay, Klavs F. Jensen, William H. Green Jr., Tommi S. Jaakkola. 2021

176 citations recorded by the source corpus.

Prediction of a molecule's 3D conformer ensemble from the molecular graph holds a key role in areas of cheminformatics and drug discovery. Existing generative models have several drawbacks including lack of modeling important molecular geometry elements (e.g. torsion angles), separate optimization stages prone to error accumulation, and the need for structure fine-tuning based on approximate classical force-fields or computationally expensive methods such as metadynamics with approximate quantum mechanics calculations at each geometry. We propose GeoMol--an end-to-end, non-autoregressive and SE(3)-invariant machine learning approach to generate distributions of low-energy molecular 3D conformers. Leveraging the power of message passing neural networks (MPNNs) to capture local and global graph information, we predict local atomic 3D structures and torsion angles, avoiding unnecessary over-parameterization of the geometric degrees of freedom (e.g. one angle per non-terminal bond). Such local predictions suffice both for the training loss computation, as well as for the full deterministic conformer assembly (at test time). We devise a non-adversarial optimal transport based loss function to promote diverse conformer generation. GeoMol predominantly outperforms popular open-source, commercial, or state-of-the-art machine learning (ML) models, while achieving significant speed-ups. We expect such differentiable 3D structure generators to significantly impact molecular modeling and related applications.

Subfields: Computational Theory and Mathematics, Materials Chemistry, Molecular Biology

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

In Materials Chemistry

In Molecular Biology

Citations in the corpus

EquiBind: Geometric Deep Learning for Drug Binding Structure Prediction (2022, methodology): “Similar to (Ganea et al., 2021a), an important geometric induc-E tive bias is to predict the same binding complex no matter how the initial molecules are positioned and oriented in space.”
GeoDiff: a Geometric Diffusion Model for Molecular Conformation Generation (2022, methodology): “Ganea et al. (2021) studied generating conformations by predicting both bond lengths and angles.”
Torsional Diffusion for Molecular Conformer Generation (2022, methodology): “A number of machine learning methods for conformer generation has been developed [Xu et al., 2021a,b, Shi et al., 2021, Luo et al., 2021], the most recent and advanced of which are GeoMol [Ganea et al., 2021] and GeoDiff [Xu et al., 2022].”
Equivariant Diffusion for Molecule Generation in 3D (2022, background): “, 2019), with adaptions for Euclidean symmetries in (Köhler et al., 2020; Xu et al., 2021a; Simm et al., 2021; Ganea et al., 2021; Guan et al., 2022) resulting in performance improvements.”
MolGenSurvey: A Systematic Survey in Machine Learning Models for Molecule Design (2022, background): “GeoMol [171] presents a generative learning pipeline for this task which ﬁrst predicts local atomic 3D structures and torsion angles and then assembles the conformations, to avoid unnecessary over-parameterization of the geometric degrees of freedom.”
Crystal Diffusion Variational Autoencoder for Periodic Material Generation (2021, methodology): “Inspired by Xu et al. (2021a); Ganea et al. (2021), we deﬁne two coverage metrics, COV-R (Recall) and COV-P (Precision), to measure the similarity between ensembles of generated materials and ground truth materials in test set.”
Independent SE(3)-Equivariant Models for End-to-End Rigid Protein Docking (2021, background): “…constrained models ease optimization and have shown important improvements in biology or chemistry – e.g. for molecular structures (Fuchs et al., 2020; Hutchinson et al., 2020; Wu et al., 2021; Jumper et al., 2021; Ganea et al., 2021) and different types of 3D point clouds (Thomas et al., 2018).”
Generating Molecular Conformer Fields (2023, methodology): “These approaches tend to use equivariant network architectures to deal with molecular graphs (Xu et al., 2022) or model domain-specific factors like torsional angles (Ganea et al., 2021; Jing et al., 2022).”
Uni-Mol: A Universal 3D Molecular Representation Learning Framework (2023, background): “Therefore, there are many recent works that train the model to fast generate conformations from molecular graph [46, 47, 48, 49].”
Predicting Molecular Conformation via Dynamic Graph Score Matching (2021, background): “Two concurrent works [12, 44] exist which generate conformations in end-to-end fashion via geometry elements assembly and bilevel programming respectively.”
Long Range Graph Benchmark (2022, unknown): “To adapt these tasks to the general graph learning setting, we avoided the prediction of pairwise node distances, which would require specialized methods from the conformer generation literature [21, 52, 63].”