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

Abstract

Accurate prediction of protein-ligand binding affinity is pivotal for drug design and discovery. Here, we proposed a novel deep fusion graph neural networks framework named FGNN to learn the protein−ligand...

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

“…Building on this, subsequent studies have integrated both refined and general datasets. 54,109,110,128,141 Consistently with the previous results on the positive data in structure-based deep learning, Francoeur et al 54 observed a notable performance boost in a 3D-CNN model when incorporating the general set. They also included the near-native structures obtained by redocking initialized ligand structures via universal force field (UFF), 142 as positive data.…”

“…Updated features from every set of edges are then gathered to predict PLI. While other works employed a single architecture, Dong et al 110 proposed FGNN that fused two different 3D-GNN architectures in an ensemble manner. They highlighted that their strategy could improve the prediction performance on diverse datasets in terms of the expressive power of the model.…”

SadNet: a novel multimodal fusion network for protein–ligand binding affinity prediction

“…Table 3 shows the results of various models on the PDBbind2016 core set, including benchmarks: Fast, DLSSAffinity and FGNN3. 46 The data in the table comes from the optimal model trained with multiple random seeds. It can be found from the table that compared with Pocket_net (no virtual node) which only uses pocket information, SadNet benefits from mid-term virtual node fusion and late stacking fusion to obtain global sequence representation, and the model performance is significantly improved.…”