EquiBind: Geometric Deep Learning for Drug Binding Structure Prediction

By Hannes Stärk, Octavian-Eugen Ganea, Lagnajit Pattanaik, Regina Barzilay, Tommi S. Jaakkola. 2022

380 citations recorded by the source corpus.

Predicting how a drug-like molecule binds to a specific protein target is a core problem in drug discovery. An extremely fast computational binding method would enable key applications such as fast virtual screening or drug engineering. Existing methods are computationally expensive as they rely on heavy candidate sampling coupled with scoring, ranking, and fine-tuning steps. We challenge this paradigm with EquiBind, an SE(3)-equivariant geometric deep learning model performing direct-shot prediction of both i) the receptor binding location (blind docking) and ii) the ligand's bound pose and orientation. EquiBind achieves significant speed-ups and better quality compared to traditional and recent baselines. Further, we show extra improvements when coupling it with existing fine-tuning techniques at the cost of increased running time. Finally, we propose a novel and fast fine-tuning model that adjusts torsion angles of a ligand's rotatable bonds based on closed-form global minima of the von Mises angular distance to a given input atomic point cloud, avoiding previous expensive differential evolution strategies for energy minimization.

Subfields: Computational Theory and Mathematics, Molecular Biology, Graph Learning, 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

In Molecular Biology

In Graph Learning

In Computational Chemistry & Drug Discovery

Citations in the corpus

End-to-end protein–ligand complex structure generation with diffusion-based generative models (2022, methodology): “The authors reported signiﬁcant performance gains over existing methods on the PDBbind [37] benchmark dataset and highlighted the critical issues with regression-based frameworks [38,39].”
TANKBind: Trigonometry-Aware Neural NetworKs for Drug-Protein Binding Structure Prediction (2022, methodology): “More recently, EquiBind [36] takes a new approach by directly predicting the key points on both the protein and the compound, and aligning their key points through the ingeniously designed optimal transport loss.”
DiffDock: Diffusion Steps, Twists, and Turns for Molecular Docking (2022, methodology): “EquiBind + SMINA/GNINA [3], the bounding box in which GNINA/SMINA searches for binding poses is constructed around the prediction of EquiBind with the –autobox_ligand option of GNINA/SMINA.”
Torsional Diffusion for Molecular Conformer Generation (2022, methodology): “Note that in practice, we compute an upper bound to this optimal RMSD using the fast von Mises torsion matching procedure proposed by Stärk et al. [2022].”
Dynamic-Backbone Protein-Ligand Structure Prediction with Multiscale Generative Diffusion Models (2022, methodology): “Results are compared to a recent learning-based method EquiBind [43]; for reference, we also include results from a physics-based blind docking method CB-Dock [44] obtained with ground-truth all-atom receptor inputs and using a computing budget similar to learning-based methods.”
Benchmarking AlphaFold‐enabled molecular docking predictions for antibiotic discovery (2022, methodology): “Model predictions based on the following rescoring functions were ensembled in sequence: RF-Score, NNScore, PLEC score, and RF-Score-VS.”
Do Deep Learning Models Really Outperform Traditional Approaches in Molecular Docking? (2023, methodology): “EquiBind [10] proposed a SE(3)-equivariant geometric deep learning model to predict the molecular positions and conformations by directly predicting the 3D atom coordinates of molecules.”
Artificial intelligence in drug development (2025, neutral): “AI-based receptor–ligand docking models can predict ligand spatial transformations, directly generate complex atomic coordinates using algorithms like equivariant neural networks 44–46 and learn the probability density distribution of receptor– ligand distances to generate binding poses 47–49 .”
3D Equivariant Diffusion for Target-Aware Molecule Generation and Affinity Prediction (2023, methodology): “We perform a more difﬁcult time split as St¨ark et al. (2022) in which the test set consists of structures deposited after 2019, and the training and validation set consist of earlier structures.”
Uni-Mol: A Universal 3D Molecular Representation Learning Framework (2023, background): “Equibind [78] is a recent graph deep learning based methods.”
Everything is Connected: Graph Neural Networks (2023, background): “It seems that protein folding, protein design, and protein binding prediction (Stärk et al., 2022) all appear to be an extremely potent area of attack for geometric GNNs; just one of many solid reasons why the field of structural biology would benefit from these recent developments (Bouatta et…”
MolGenSurvey: A Systematic Survey in Machine Learning Models for Molecule Design (2022, background): “EquiBind [184] instead represents molecules and protein pockets with 3D geometries and models them with a 3D equivariant graph neural network.”
Structure-based drug design with equivariant diffusion models (2022, positive): “, 2020), binding pose prediction (Stärk et al., 2022) or molecular dynamics (Batzner et al.”