Accurate and interpretable nanoSAR models from genetic programming-based decision tree construction approaches.

Area of Science:

• Nanomaterial safety
• Computational toxicology
• Environmental health

Background:

• Engineered nanomaterials (ENMs) offer novel properties but pose potential health and environmental risks.
• Existing structure-activity relationship (SAR) models for nanotoxicity (nanoSAR) have limitations in robustness, predictivity, and interpretability, especially with sparse data.
• Current SAR methods often require large datasets, struggle with feature selection, and may not capture nonlinear relationships.

Purpose of the Study:

• To introduce and evaluate a novel algorithm, genetic programming-based decision tree construction (GPTree), for nanoSAR modelling.
• To demonstrate GPTree's ability to construct accurate and interpretable nanoSAR models.
• To overcome the limitations of traditional SAR methods in handling sparse and complex nanotoxicity data.

Main Methods:

• Application of a genetic programming-based decision tree construction tool (GPTree) to nanoSAR modelling.
• Utilizing GPTree to analyze four diverse literature datasets on ENM biological effects.
• Comparing GPTree's model performance against existing methods on the same datasets.

Main Results:

• GPTree successfully constructed accurate and interpretable nanoSAR models across four diverse datasets.
• Models generated by GPTree demonstrated equivalent or superior accuracy compared to previous studies.
• GPTree automatically selects relevant descriptors and handles small datasets effectively.

Conclusions:

• GPTree is a robust and automatic method for developing accurate nanoSAR models.
• GPTree offers significant advantages in interpretability and performance, particularly with limited data.
• This approach enhances the ability to predict and minimize risks associated with engineered nanomaterials.