hotWater

hotWater predicts water binding sites ("water hot-spots") on protein surfaces using a residual deep learning model trained on RCSB PDB data to identify regions of high water affinity relevant to protein hydration and ligand interactions.

Key Features:

• Deep Learning Model: A residual deep learning architecture trained on extensive RCSB PDB data to predict potential water binding sites.
• Efficiency: Provides a computationally efficient alternative to molecular dynamics simulations for identifying water hot-spots.
• Versatile Input and Outputs: Accepts a PDB file or accession code and can rank existing water molecules by predicted binding strength or scan protein surfaces de novo to identify likely water hot-spots.
• Cross-Resolution Capability: Validated across high-resolution X-ray crystallography and lower-resolution techniques such as NMR and CryoEM, and able to recover known water binding sites from "water-free" structures.
• Performance vs Electron-Density Methods: Surpasses traditional algorithms that rely solely on electron density, with higher recall particularly at resolutions greater than 2.6 Å.

Scientific Applications:

• Structural Biology: Identifies water hot-spots to aid interpretation of protein hydration dynamics and their implications for protein function.
• Drug Discovery: Informs drug design by highlighting potential interaction points on target proteins, including applications to the RNA polymerase complex from SARS-CoV-2.

Methodology:

Uses a residual deep learning architecture trained on high-resolution X-ray diffraction structures from the RCSB PDB, with electron density-indicated water positions from X-ray data used as training labels.