Related Experiment Videos
Comparison of correlation vector methods for ligand-based similarity searching
Uli Fechner1, Lutz Franke, Steffen Renner
1Johann Wolfgang Goethe-Universität, Institut für Organische Chemie und Chemische Biologie, Marie-Curie-Str. 11, D-60439 Frankfurt, Germany.
Journal of Computer-Aided Molecular Design
|April 8, 2004
Summary
Correlation vector methods effectively screen ligands, retrieving up to 78% of active molecules. Alignment-free descriptors offer speed for large datasets, but using multiple descriptors is recommended for comprehensive ligand-based similarity searching.
Area of Science:
- Computational chemistry
- Cheminformatics
- Drug discovery
Background:
- Ligand-based virtual screening is crucial for identifying potential drug candidates.
- Efficient methods are needed to analyze large chemical databases.
Purpose of the Study:
- To evaluate correlation vector methods for ligand-based virtual screening.
- To compare different molecular descriptors and similarity measures.
Main Methods:
- Tested three alignment-free molecular descriptors (pharmacophore points, partial atom charges).
- Compared three similarity measures: Manhattan distance, Euclidean distance, Tanimoto coefficient.
- Performed retrospective analysis on a reference database.
Main Results:
- Achieved significant enrichment of active compounds.
- Retrieved up to 78% of active molecules within the top 5% of the database.
- Alignment-free descriptors showed high execution speed for coarse-grained filtering.
- Different descriptors identified distinct sets of top-ranking active molecules.
Conclusions:
- Correlation vector methods are useful for ligand-based virtual screening.
- Manhattan distance is a suitable similarity index for single-use applications.
- No single descriptor universally outperformed others; suitability depends on the specific interaction.
- Recommends using multiple descriptors in parallel for robust ligand-based similarity searching.