DeepProSite

DeepProSite predicts protein binding sites by integrating protein sequence embeddings and three-dimensional structural information to identify residues that bind proteins, peptides, nucleic acids, and other ligands.

Key Features:

• Structure Generation: Uses ESMFold to generate three-dimensional protein structures for downstream analysis.
• Sequence Representation: Derives sequence embeddings from pretrained language models to capture sequence-adjacent contextual features.
• Graph Transformer Utilization: Formulates binding site prediction as a graph node classification problem combining structural and sequence information via a Graph Transformer.
• Performance Superiority: Outperforms state-of-the-art methods on protein-protein and peptide binding site prediction benchmarks and maintains high performance on unbound structures.
• Generalization Capability: Extends predictive targets beyond proteins and peptides to nucleic acids and other ligands.

Scientific Applications:

• Structural Biology: Identifies binding residues to support elucidation of protein interactions and molecular mechanisms.
• Drug Discovery: Predicts ligand-binding residues to inform therapeutic target identification and structure-guided drug design.
• Functional Annotation: Aids annotation of protein binding sites for proteins, peptides, nucleic acids, and other ligands.

Methodology:

Protein structures are generated with ESMFold; sequence information is encoded via pretrained language models; structural and sequence features are integrated into a graph representation and classified at the node level using a Graph Transformer.