GeoBind

GeoBind predicts nucleic acid binding sites on protein surfaces by applying geometric deep learning to learn high-level surface representations from point-cloud protein models.

Key Features:

• Surface segmentation: Segments protein surfaces to enable site-level prediction of nucleic acid binding sites.
• Point-cloud input processing: Processes entire point clouds representing protein surfaces as model inputs rather than relying on handcrafted features.
• Geometric deep learning aggregation: Learns high-level representations by aggregating information from local neighbors within reference frames.
• Nucleic acid binding site prediction: Predicts nucleic acid binding interfaces on protein surfaces from learned surface representations.
• Benchmark performance: Demonstrated superior performance compared to existing state-of-the-art predictors on benchmark datasets.
• Extension to other ligands: Extended to predict various other types of ligand binding sites with competitive performance.
• Multimer and molecular surface analysis: Capable of analyzing molecular surfaces and interfaces in proteins involved in multimer formation.

Scientific Applications:

• Nucleic acid–protein interaction mapping: Identification of protein surface regions that interact with nucleic acids for studies of regulatory mechanisms.
• Interface analysis in multimers: Analysis of molecular surfaces and interfaces in proteins involved in multimer formation to investigate assembly-related functions.
• Protein–ligand interaction prediction: Prediction of diverse ligand binding sites to support biochemical and molecular biology investigations of protein–ligand interactions.

Methodology:

Segments protein surfaces, processes entire point clouds representing protein surfaces as inputs, and applies geometric deep learning that aggregates information from local neighbors within reference frames to learn high-level surface representations.