Ligand-based virtual screening by novelty detection with self-organizing maps.

Area of Science:

• Computational Chemistry
• Cheminformatics
• Drug Discovery

Background:

• Ligand-based virtual screening aims to identify novel bioactive compounds.
• Existing methods often require known inactive compounds, which are scarce in databases.
• Novelty detection offers a way to identify compounds outside the known activity space.

Purpose of the Study:

• To present a novel ligand-based virtual screening method using Self-Organizing Maps (SOM) for novelty detection.
• To address the limitation of scarce inactive compounds in virtual screening.
• To evaluate the performance of SOM-based novelty detection against traditional similarity searches.

Main Methods:

• Utilized Self-Organizing Maps (SOM) as a novelty detection tool for virtual screening.
• Represented chemical structures using spatial autocorrelation functions weighted by atomic physicochemical properties.
• Compared SOM-based novelty detection with similarity searches using Daylight fingerprints and data fusion.

Main Results:

• SOM-based novelty detection achieved significant enrichment factors (105-462) in retrospective screening.
• This method demonstrated a 25%-100% improvement over Daylight fingerprint similarity searches for top-ranked compounds.
• The two methods, novelty detection and similarity search, were found to be complementary.

Conclusions:

• Novelty detection with SOM is a valuable tool for improving the retrieval of potentially active compounds.
• It can be used in conjunction with other virtual screening methods.
• SOM-based novelty detection serves as an effective library design tool and for selecting promising compounds from large datasets.