Safety-by-design using forward and inverse multi-target machine learning.

Area of Science:

• Nanotechnology
• Materials Science
• Computational Chemistry

Background:

• Nanomaterial safety is crucial for sustainable nanotechnology development.
• Current safety-by-design frameworks face challenges in balancing multiple nanoparticle criteria.
• Predicting and mitigating potential toxicity of nanomaterials before use is essential.

Purpose of the Study:

• To develop a multi-target machine learning model for predicting titania nanoparticle characteristics in sunscreens.
• To establish direct structure/product relationships for optimizing multiple design criteria simultaneously.
• To create inverse design models for identifying optimal nanoparticle configurations meeting safety and performance thresholds.

Main Methods:

• Utilized a synthetic dataset of over 19,000 sunscreen product specifications.
• Employed multi-target machine learning to predict nanoparticle size, shape, concentration, and polytype.
• Inverted, re-optimized, and re-trained models for inverse design capabilities.

Main Results:

• Demonstrated superior performance of direct structure/product relationships over conventional methods for simultaneous property prediction.
• Successfully predicted nanoparticle configurations meeting sun protection, transparency, and toxicity requirements.
• Identified novel nanoparticle structures outside the initial training set.

Conclusions:

• Directly predicting nanoparticle structures for multiple performance and safety criteria offers a novel safety-by-design approach.
• This methodology can be applied to diverse products and materials requiring simultaneous optimization of multiple design parameters.
• Machine learning enables proactive identification of safer nanomaterial formulations, supporting responsible innovation.