Geometry-complete perceptron networks for 3D molecular graphs

Alex Morehead,Jianlin Cheng

DOI: https://doi.org/10.1093/bioinformatics/btae087

IF: 5.8

2024-02-01

Bioinformatics

Abstract:Abstract Motivation The field of geometric deep learning has recently had a profound impact on several scientific domains such as protein structure prediction and design, leading to methodological advancements within and outside of the realm of traditional machine learning. Within this spirit, in this work, we introduce GCPNet, a new chirality-aware SE(3)-equivariant graph neural network designed for representation learning of 3D biomolecular graphs. We show that GCPNet, unlike previous representation learning methods for 3D biomolecules, is widely applicable to a variety of invariant or equivariant node-level, edge-level, and graph-level tasks on biomolecular structures while being able to (1) learn important chiral properties of 3D molecules and (2) detect external force fields. Results Across four distinct molecular-geometric tasks, we demonstrate that GCPNet’s predictions (1) for protein–ligand binding affinity achieve a statistically significant correlation of 0.608, more than 5%, greater than current state-of-the-art methods; (2) for protein structure ranking achieve statistically significant target-local and dataset-global correlations of 0.616 and 0.871, respectively; (3) for Newtownian many-body systems modeling achieve a task-averaged mean squared error less than 0.01, more than 15% better than current methods; and (4) for molecular chirality recognition achieve a state-of-the-art prediction accuracy of 98.7%, better than any other machine learning method to date. Availability and implementation The source code, data, and instructions to train new models or reproduce our results are freely available at https://github.com/BioinfoMachineLearning/GCPNet.

biochemical research methods,biotechnology & applied microbiology,mathematical & computational biology

What problem does this paper attempt to address?

The paper attempts to address the following issues: 1. **Molecular Chirality Recognition**: A new geometric deep learning model (GCPN ET) is proposed, which can effectively recognize the chiral characteristics of 3D molecules. Chirality refers to the property of a molecule that cannot be superimposed on its mirror image through rotation and translation, which is crucial in fields such as drug design. 2. **Protein-Ligand Binding Affinity Prediction**: Using 3D molecular graph representation, the binding affinity between proteins and ligands is predicted. This task is of great significance for rapid screening in the drug discovery process. 3. **Protein Structure Ranking**: Evaluating the quality of a given protein structure and comparing it with a reference structure. This is important when designing drugs for specific protein targets, especially when the 3D structure of these targets has not been experimentally determined. 4. **Newtonian Multi-body System Modeling**: Simulating atomic systems in the real world, detecting and utilizing the force field information within them. This is significant for understanding complex biomolecular systems. The main contribution of the paper is the proposal of a novel geometric perception neural network architecture, GCPN ET. This model not only possesses SE(3) equivariance but also sensitively captures the chiral characteristics of molecules and effectively detects the global physical forces acting on each atom. Additionally, GCPN ET achieves significantly better results than existing methods on four different molecular geometry tasks.

Geometry-complete perceptron networks for 3D molecular graphs

Leveraging 2D molecular graph pretraining for improved 3D conformer generation with graph neural networks

3D Molecular Geometry Analysis with 2D Graphs

GeomGCL: Geometric Graph Contrastive Learning for Molecular Property Prediction

A Universal Framework for Accurate and Efficient Geometric Deep Learning of Molecular Systems

Molecular geometric deep learning

Graph neural network for 3-dimensional structures including dihedral angles for molecular property prediction

Learning 3D Representations of Molecular Chirality with Invariance to Bond Rotations

Geometric Graph Representations and Geometric Graph Convolutions for Deep Learning on Three-Dimensional (3D) Graphs

GTAM: A Molecular Pretraining Model with Geometric Triangle Awareness

Learning Universal and Robust 3D Molecular Representations with Graph Convolutional Networks

ChemRL-GEM: Geometry Enhanced Molecular Representation Learning for Property Prediction

ChiENN: Embracing Molecular Chirality with Graph Neural Networks

3DMol-Net: Learn 3D Molecular Representation using Adaptive Graph Convolutional Network Based on Rotation Invariance

Curvature-enhanced Graph Convolutional Network for Biomolecular Interaction Prediction

3DReact: Geometric deep learning for chemical reactions

Molecule3D: A Benchmark for Predicting 3D Geometries from Molecular Graphs

ComENet: Towards Complete and Efficient Message Passing for 3D Molecular Graphs

Pre-training Molecular Graph Representation with 3D Geometry

Non-equilibrium molecular geometries in graph neural networks