Machine Learning Allowed Interpreting Toxicity of a Fe-Doped CuO NM Library Large Data Set─An Environmental In Vivo...

Area of Science:

• Environmental Toxicology
• Nanomaterial Science
• Computational Toxicology

Background:

• Nanomaterial (NM) variability complicates safety assessments.
• Predictive modeling is essential for understanding NM impacts on organisms.
• Existing ecotoxicological guidelines may not capture long-term effects.

Purpose of the Study:

• To investigate the ecotoxicity of custom-designed iron-doped copper oxide (Fe-CuO) nanomaterials.
• To utilize machine learning (ML) to identify key nanomaterial characteristics driving toxicity.
• To compare short-term (21-day) versus long-term (49-day) exposure effects on soil invertebrates.

Main Methods:

• Ecotoxicological testing of 9 Fe-doped CuO NMs using the soil invertebrate Enchytraeus crypticus.
• Application of machine learning models to a dataset including 68 descriptors, 6 concentrations, and 2 exposure times.
• Analysis of both environmental parameters (e.g., zeta potential) and particle-specific descriptors.

Main Results:

• 10% Fe-CuO was the most toxic NM; Fe3O4 NM was the least toxic.
• Environmental parameters were key predictors in short-term (21-day) tests.
• Particle-specific descriptors became more important in longer-term (49-day) exposures, revealing decreased concentration-response steepness with higher Fe content.

Conclusions:

• Longer-term exposure studies (49 days) are crucial for accurate nanomaterial hazard assessment, complementing standard OECD guidelines.
• Machine learning effectively links nanomaterial descriptors to ecotoxicological effects, outperforming traditional modeling.
• Particle characteristics and longer exposure durations are critical factors in understanding nanomaterial environmental risks.