GalaxyWater-CNN

GalaxyWater-CNN predicts positions of water molecules on protein structures using a 3D-convolutional neural network to generate water score maps that inform water-mediated protein structure-function relationships.

Key Features:

• 3D-convolutional neural network (3D-CNN): A 3D-CNN model is trained to generate a water score map for given protein structures.
• Training dataset: Trained on high-resolution protein crystal structures that include resolved water molecules.
• Prediction scope: Predicts water positions along protein chains, at protein-protein interfaces, and within protein-compound binding sites.
• Structural-data approach: Leverages structural data rather than conventional energy functions for water placement.
• Performance: Achieves superior coverage of crystal water molecules and improved precision of predicted water positions compared to energy-based methods.
• Hydrogen-bond network prediction: Accurately predicts water molecules involved in forming hydrogen-bond networks.

Scientific Applications:

• Protein structure-function analysis: Enhances understanding of protein structure-function relationships where explicit waters are biologically relevant.
• Protein-compound binding studies: Informs placement of water molecules within protein-compound binding sites for binding-site characterization.
• Protein-protein interface analysis: Identifies interfacial waters that mediate protein-protein interactions.
• Interpretation of crystal-resolved waters: Improves coverage and precision in mapping experimentally observed water molecules from crystal structures.

Methodology:

Employs a 3D-CNN trained on high-resolution protein crystal structures with resolved water molecules to produce water score maps, using structural data rather than conventional energy functions.