Self-organizing maps and molecular similarity

Area of Science:

• Computational chemistry
• Cheminformatics
• Artificial intelligence in drug discovery

Background:

• Self-organizing maps (SOMs) reduce data dimensionality.
• Kohonen neural networks are effective for pattern recognition.
• Previous studies classified muscarinic/nicotinic agonists using SOMs.

Purpose of the Study:

• To apply Kohonen networks for analyzing histamine H2 agonists.
• To generate 2D representations of electrostatic potentials.
• To rank histamine H2 agonists by biological activity.

Main Methods:

• Utilizing Kohonen neural networks for dimensionality reduction.
• Generating two-dimensional maps of electrostatic potential.
• Analyzing the ring structures of histamine H2 agonists.

Main Results:

• Successful generation of 2D representations of electrostatic potential.
• Demonstrated the method's extension to ranking drug candidates.
• Provided a novel approach for predicting histamine H2 agonist activity.

Conclusions:

• Kohonen networks offer a valuable tool for drug discovery.
• The method effectively visualizes and ranks molecules by activity.
• This approach advances the understanding of structure-activity relationships.