Geometric Interaction Graph Neural Network for Predicting Protein–Ligand Binding Affinities from 3D Structures (GIGN)

Abstract

Predicting protein–ligand binding affinities (PLAs) is a core problem in drug discovery. Recent advances have shown great potential in applying machine learning (ML) for PLA prediction. However, most of them omit the 3D structures of complexes and physical interactions between proteins and ligands, which are considered essential to understanding the binding mechanism. This paper proposes a geometric interaction graph neural network (GIGN) that incorporates 3D structures and physical interactions for predicting protein–ligand binding affinities. Specifically, we design a heterogeneous interaction layer that unifies covalent and noncovalent interactions into the message passing phase to learn node representations more effectively. The heterogeneous interaction layer also follows fundamental biological laws, including invariance to translations and rotations of the complexes, thus avoiding expensive data augmentation strategies. GIGN achieves state-of-the-art performance on three external test sets. Moreover, by visualizing learned representations of protein–ligand complexes, we show that the predictions of GIGN are biologically meaningful.

Ligand binding affinity prediction with fusion of graph neural networks and 3D structure-based complex graph

“…In terms of the models trained on the PDBbind refined set, TopBP 75 ( R p = 0.861) is the best one based on CNN and MP-GNN 76 ( R p = 0.852) is the best one based on GNN. Other methods, such as 3D CNN and GNN based HAC-Net, 77 the graph-based DL model GIGN, 78 InteractionGraphNet 79 and PLANET, 80 the topological fingerprints-based model (TNet-Bp 81 ), and the 3D CNN-based model (AK-Score 43 ), also show good scoring power in the CASF-2016 test. Other ML-based methods also perform good, for example PerSpect ML 82 ( R p = 0.84, RMSE = 1.27).…”

Toward generalizable structure‐based deep learning models for protein–ligand interaction prediction: Challenges and strategies

“…64,116,117 EGNN kept the rototranslational equivariance for both output and hidden layers and thus provide a more powerful representation for 3D protein-ligand structures, leading to higher performance in both prediction and generalizability. Although the original model was not designed for PLI prediction, Yang et al 112 assessed EGNN on the PDBbind core sets and CSAR benchmark, together with GIGN (see Table 2). To capture both the coarser and finer details of proteins, authors of HOLOPROT 118 proposed a novel multi-scale graph representation consisting of surface and structure-level graphs.…”

“…Many proposed PLI models employed the grid search method to find the best hyperparameter combination. 66,69,111,112 While this approach ensures complete exploration of the hyperparameter space, it has the drawback of exponential computational complexity as the number of hyperparameters increases. To avoid such a high computational cost, several works tuned each hyperparameter one at a time in a pre-defined search space.…”

Exploring protein–ligand binding affinity prediction with electron density-based geometric deep learning

“…Aiming to limit the number of laboratory experiments and enable more rapid ligand design, several computational approaches have been developed for in silico binding affinity prediction. 2 Structure-based deep learning has received particular attention for this specific task, 3–5 as well as for binding site identification, 6,7 molecular docking, 8,9 and de novo molecular design. 10…”

“…In addition to geometric deep learning approaches, 5,23–26 also simulation-based approaches ( e.g. , free energy perturbation, 27,28 MM/PBSA 29–31 ) are frequently used for binding affinity prediction.…”