Machine Learning on Systematically Curated Data Reveals Key Determinants of Magnetic Hyperthermia Performance.

Area of Science:

• Biomedical Engineering
• Materials Science
• Computational Science

Background:

• Accurate specific absorption rate (SAR) prediction is crucial for superparamagnetic iron oxide nanoparticles (SPIONs) in magnetic hyperthermia.
• Optimizing SPIONs requires understanding their properties and experimental parameters influencing SAR.

Purpose of the Study:

• To develop and validate a machine learning model for predicting SPION SAR.
• To identify key features influencing SPION SAR for optimized nanoparticle design.

Main Methods:

• A comprehensive dataset of 1850 entries from 84 studies was curated, encompassing 30 predictive features.
• Twelve machine learning algorithms were evaluated, with CatBoost optimized using Bayesian hyperparameter tuning.
• Shapley additive explanation and conformal prediction were employed for analysis and reliability assessment.

Main Results:

• The CatBoost model achieved a high R² of 0.98, demonstrating excellent predictive accuracy.
• Alternating magnetic field amplitude and frequency were identified as the most significant factors affecting SAR.
• The model showed strong predictive performance for smaller SPIONs (≈7 nm), with increased variability for larger particles.

Conclusions:

• Advanced machine learning models enable precise SAR prediction for SPIONs.
• The study provides critical insights into SPION design for enhanced magnetic hyperthermia applications.
• This work supports the systematic optimization of SPIONs for clinical translation in cancer therapy.