Hybrid protein-ligand binding residue prediction with protein language models: Does the structure matter?

Abstract

AbstractPredicting protein-ligand binding sites is crucial in studying protein interactions with applications in biotechnology and drug discovery. Two distinct paradigms have emerged for this purpose: sequence-based methods, which leverage protein sequence information, and structure-based methods, which rely on the three-dimensional (3D) structure of the protein. To enhance the state-of-the-art performance in this field, we propose a novel approach combining both paradigms’ strengths. Our hybrid model integrates two recent deep learning architectures: protein language models (pLMs) from the sequence-based paradigm and Graph Neural Networks (GNNs) from the structure-based paradigm. Specifically, we construct a residue-level Graph Attention Network (GAT) model based on the protein’s 3D structure that uses pre-trained pLM embeddings as node features. This integration enables our model to capture both the sequential information encoded in the protein sequence and the structural relationships within the protein. The model has improved state-of-the-art performance on a benchmark dataset over a range of ligands and ligand types. Ablation studies have demonstrated the role of the graph attention mechanism, particularly in densely connected graphs. Moreover, we have shown that as more complex pLMs are employed to represent node features, the relative impact of the GNN architecture diminishes. This observation suggests that, to some extent, the structural information required for accurate binding site prediction is inherently captured by the pLMs themselves. protein-ligand binding sites, binding residues prediction, graph neural networks, graph attention, protein language models, protein embeddings