Rational prediction of the performance of pH-responsive functionalized iron oxide grafted on graphene oxide for magnetic hyperthermia cancer therapy using machine learning

Cancer remains a major global health challenge, demanding innovative treatment strategies with minimal side effects, in line with the 2030 Sustainable Development Goals (SDGs). Magnetic hyperthermia therapy (MHT) has emerged as a promising approach, yet its realization requires exploration of predictive tools capable of capturing complex interactions between nanoparticle properties and tumor microenvironments. This study explores the application of machine learning (ML) to predict the performance of pH-responsive graphene oxide-modified iron oxide nanoparticles forming magnetic hybrid nanostructures (MHNS). Two ANFIS ML approaches, subtractive clustering and grid partitioning, were implemented to model heating performance as a function of time, composition, and pH. The influence of key hyperparameters on prediction accuracy was systematically analyzed to demystify model behavior, and both raw and transformed datasets were evaluated. For benchmarking, additional ML models were also implemented, and feature sensitivity analysis was conducted to quantify the relative influence of input variables. Results show that ANFIS with subtractive clustering using the raw dataset achieved the highest accuracy (testing R² = 0.9938) with negligible error metrics. Feature analysis highlighted time and pH as dominant predictors, underscoring the practical relevance of acidic tumor microenvironments. This study provides the first demonstration of accurate ML-based prediction and demystification of pH-responsive MHNS performance in MHT, enabling accelerated and efficient optimization of cancer treatment strategies with minimal side effects.