GraphSite

GraphSite predicts protein-DNA binding sites by integrating graph transformer models with AlphaFold2 structural predictions to classify DNA-binding residues from protein sequences.

Key Features:

• Integration with AlphaFold2: GraphSite incorporates AlphaFold2-predicted protein structures to provide three-dimensional spatial information for binding-site prediction.
• Graph Transformer Model: It formulates prediction as a graph node classification task and applies a transformer-based graph model to capture both local and global structural context.
• Performance metrics: On an independent test of 181 proteins, GraphSite increased the area under the precision-recall curve by 16.4% and the Matthews correlation coefficient by 11.2% compared to existing structure-based methods.

Scientific Applications:

• Mechanistic analysis: Enables identification of DNA-binding residues to elucidate mechanistic bases of protein-DNA interactions involved in transcription, DNA repair, and other processes.
• Therapeutic design: Supports design of therapeutic agents targeting specific protein-DNA complexes by pinpointing interface residues.

Methodology:

Protein structural data are converted into a graph where nodes represent amino acids and edges capture spatial relationships informed by AlphaFold2 predictions; a transformer-based graph model processes this graph to classify nodes as DNA-binding or non-binding residues using sequence and structural information.