Transcending Structural Dependencies: A Tunable Mass Spectrometry-Driven Machine Learning Framework for Genotoxicity...

Area of Science:

• Analytical Chemistry
• Toxicology
• Computational Chemistry

Background:

• Genotoxic compounds (GCs) pose significant risks in environmental, pharmaceutical, and food safety.
• Current analytical methods often require prior structural information or are time-consuming.
• There is a need for efficient, structure-independent methods for GC detection.

Purpose of the Study:

• To develop and validate GenoToxMass, a mass spectrometry-based predictive model for genotoxic compounds.
• To eliminate the dependence on prior chemical structure information for GC screening.
• To provide a scalable and accessible tool for regulatory agencies and researchers.

Main Methods:

• Development of a predictive model using a dataset of 16,806 mass spectra.
• Utilization of multimodal features for model construction.
• Validation using external datasets and real-world applications (Ames test concordance).

Main Results:

• GenoToxMass achieved an AUC of 0.95, with external validation AUC of 0.89.
• Demonstrated >85% concordance with Ames test outcomes and literature data.
• Successfully classified 42 chemical products and 16 drug impurities in complex matrices.

Conclusions:

• GenoToxMass provides a versatile, scalable paradigm for non-targeted GC screening.
• The model offers transparency through interpretable spectral features, diverging from traditional toxicological methods.
• An accessible web platform and SOP facilitate practical deployment for regulatory and industrial use.