RDKit

RDKit implements a fragment-based coordinate generation method for molecular structure prediction, producing three-dimensional coordinates with improved stereochemical accuracy for use with Open Babel.

Key Features:

• Fragment-based coordinate generation: Generates 3D coordinates by assembling precomputed fragments rather than relying solely on rules-based or distance-geometry approaches.
• Open Babel integration: Introduces the fragment-based coordinate generation method specifically for use with the Open Babel toolkit.
• Fragment decomposition at rotatable bonds: Decomposes input molecules into fragments by cutting at rotatable bonds.
• Fragment library from open crystallographic databases: Assigns fragment coordinates using a comprehensive fragment library derived from open crystallographic databases.
• Simultaneous multi-atom coordinate determination: Determines multiple atomic coordinates at once to accelerate structure prediction compared to traditional implementations.
• Improved stereochemical accuracy: Enhances stereochemical correctness of generated geometries.
• Preservation of bond geometry: Maintains or improves precision of bond lengths, bond angles, and dihedral torsions.
• Performance vs rules-based and distance-geometry methods: Delivers increased speed relative to rules-based implementations and distance geometry methods used in RDKit.

Scientific Applications:

• Computational property prediction for polymers: Provides 3D geometries used in property prediction workflows for polymers.
• Molecular materials modeling: Supplies accurate molecular geometries for simulation and modeling of molecular materials.
• Drug design and cheminformatics: Generates stereochemically accurate structures for structure-based drug design and related cheminformatics analyses.
• Reproducible structure prediction: Enables reproducible generation of reliable three-dimensional molecular geometries for computational studies.

Methodology:

Input molecules are decomposed at rotatable bonds into fragments, fragment coordinates are assigned from a library derived from open crystallographic databases, and multiple atomic coordinates are determined simultaneously to accelerate prediction compared to rules-based or distance-geometry methods.