Drug/nondrug classification with consensual Self-Organising Map and Self-Organising Global Ranking algorithms

Area of Science:

• Computational chemistry
• Cheminformatics
• Machine learning in drug discovery

Background:

• Accurate classification of druglike compounds is crucial for efficient drug discovery.
• Existing methods like Self Organising Map (SOM) have limitations.
• A consensus approach can enhance classification reliability.

Purpose of the Study:

• To develop and evaluate a special consensual approach for separating druglike from non-druglike compounds.
• To improve classification accuracy compared to individual algorithms.
• To leverage group decision-making for robust predictions.

Main Methods:

• A consensual model integrating Self Organising Global Ranking (SOGR) and Self Organising Map (SOM) algorithms.
• Preprocessing involves random matrix transformation and median filtering.
• Postprocessing utilizes a consensus unit for combining individual classifications.
• Focus on a novel neighborhood concept differentiating from standard SOM.

Main Results:

• The consensual model achieved 90.63% accuracy for classifying confirmed drugs.
• Non-drug compounds were classified with 80.44% accuracy.
• Self Organising Global Ranking (SOGR) demonstrated superior performance over the Self Organising Map (SOM) algorithm.

Conclusions:

• The proposed consensual approach significantly enhances the accuracy of druglikeness prediction.
• Combining multiple classification results via consensus offers a robust strategy.
• SOGR-based consensus shows promise for improved cheminformatics applications.