Rational Design of Safer Inorganic Nanoparticles via Mechanistic Modeling-informed Machine Learning.

Area of Science:

• Nanotechnology
• Toxicology
• Computational Biology

Background:

• Assessing inorganic nanoparticle (NP) safety is crucial for clinical applications.
• Existing methods for NP toxicity evaluation are often time-consuming and resource-intensive.

Purpose of the Study:

• To develop a machine-learning (ML) framework for predicting NP toxicity.
• To enable accurate in vitro and in vivo NP safety assessments.
• To facilitate the rational design of safer nanomaterials for clinical translation.

Main Methods:

• Developed an ML framework utilizing physicochemical properties and experimental conditions to predict NP toxicity.
• Trained and validated binary classification models using an in vitro cytotoxicity dataset.
• Integrated a physiologically-based pharmacokinetic (PBPK) model to quantify organ-specific NP exposure for in vivo predictions.
• Validated the framework's predictive accuracy using external datasets and mesoporous silica NPs.

Main Results:

• The ML framework accurately predicted in vitro NP cytotoxicity.
• Explainability analysis identified key determinants of NP toxicity and established structure-toxicity relationships.
• PBPK-informed ML models demonstrated robust predictions of organ-specific nanotoxicity in vivo.
• The developed approach serves as a Novel Alternative Method (NAM) for streamlined NP safety assessment.

Conclusions:

• The PBPK-informed ML framework offers a powerful tool for predicting NP toxicity.
• This approach can significantly accelerate the safety evaluation of nanomaterials.
• Enables the design of safer NPs, promoting their clinical translation and therapeutic applications.