Unlocking the Potential of Clustering and Classification Approaches: Navigating Supervised and Unsupervised Chemical...

Area of Science:

• Toxicology and computational chemistry
• Development and application of New Approach Methodologies (NAMs)

Background:

• Toxicology is advancing with NAMs for chemical toxicity prediction.
• Class-based methods like clustering and classification are vital for NAMs.
• Computational chemistry, data availability, and machine learning enhance these methods.

Purpose of the Study:

• To deepen the understanding of class-based approaches in toxicology.
• To elucidate the role of chemical similarity (structural and biological) in clustering and classification approaches (CCAs).
• To highlight the nuances, applications, and common misuses of CCAs.

Main Methods:

• Analysis of unsupervised (endpoint-agnostic) and supervised (endpoint-specific) similarity measures.
• Examination of how chemical representations and biological activity labels influence similarity definitions.
• Discussion of the implications of choosing appropriate similarity measures for CCAs.

Main Results:

• The effectiveness of CCAs hinges on context-dependent similarity definitions.
• Unsupervised methods use endpoint-agnostic similarity for pattern discovery.
• Supervised methods require endpoint-specific similarity for classification and prediction.

Conclusions:

• Careful selection and understanding of similarity are imperative for effective CCAs in toxicology.
• Misapplication of unsupervised methods in specific contexts, like read-across, can lead to errors.
• Distinguishing between endpoint-agnostic and endpoint-specific similarity is vital for accurate toxicological assessments.