PeakProbe

PeakProbe classifies multi-atom solvent species (e.g., sulfate) and water molecules in macromolecular X-ray crystal structures to improve automated solvent model building and validation.

Key Features:

• Resolution-independent classification: Maps 19 resolution-dependent electron density features and two chemical environment features into a two-dimensional score space to enable classification independent of resolution.
• High accuracy: Classifies peaks into four distinct solvent classes with greater than 99% accuracy on peaks derived from existing atomic coordinates and difference density maxima.
• Solvent differentiation: Distinguishes water molecules from multi-atom solvent species such as sulfate.
• Error identification: Detects peaks associated with model errors and clusters likely corresponding to multi-atom solvents.
• Model validation: Validates existing solvent models using solvent-omit electron-density maps.
• Computational integration and input formats: Operates within the phenix.python environment leveraging cctbx libraries and accepts PDB files and structure factor data as input.

Scientific Applications:

• Refinement of macromolecular crystal structures: Improves automated solvent model building and refinement by assigning solvent species to electron density peaks.
• Structural interpretation: Enhances precision of structural interpretations for studies of molecular interactions and functions.
• Drug design: Improves the quality of structural models used in drug design by distinguishing waters from multi-atom solvents that can affect binding sites.
• Enzyme mechanism studies: Facilitates enzyme mechanism analyses by refining solvent placement that influences active-site chemistry.
• Protein–ligand interaction analyses: Supports interpretation of protein–ligand interactions by validating solvent models that can alter ligand-binding conclusions.

Methodology:

Maps 19 resolution-dependent electron density features and two chemical environment features into a two-dimensional score space and uses a classifier trained on large-scale sampling of solvent models from the Protein Data Bank (PDB) to estimate relative frequencies of four solvent classes and classify peaks from existing atomic coordinates and difference density maxima, with validation using solvent-omit electron-density maps.