Predicting Dose-Dependent Carcinogenicity of Chemical Mixtures Using a Novel Hybrid Neural Network Framework and Mathematical Approach

This study addresses the challenge of assessing carcinogenic potential hazardous chemical mixtures, such as per- and polyfluorinated substances (PFASs), which are known to contribute significantly cancer development. Here, we propose a novel framework called HNNMixCancer that utilizes hybrid neural network (HNN) integrated into machine-learning framework. incorporates mathematical model simulate enabling creation classification models for binary (carcinogenic or noncarcinogenic) multiclass (categorical carcinogenicity) regression potency). Through extensive experimentation, demonstrate our HNN outperforms other methodologies, including random forest, bootstrap aggregating, adaptive boosting, support vector regressor, gradient kernel ridge, decision tree with AdaBoost, KNeighbors, achieving superior accuracy 92.7% in classification. To address limited availability experimental data enrich training data, generate an assumption-based virtual library mixtures using noncarcinogenic single all models. Remarkably, this case, methods achieve accuracies exceeding 98% In external validation tests, method achieves highest 80.5%. Furthermore, classification, demonstrates overall 96.3%, outperforming RF, Bagging, achieved 91.4%, 91.7%, 80.2%, respectively. models, HNN, SVR, GB, KR, DT KN average R2 values 0.96, 0.90, 0.77, 0.94, 0.97, respectively, showcasing their effectiveness predicting concentration at mixture becomes carcinogenic. Our exhibits exceptional predictive power prioritizing even when relying on mixtures. capability is particularly valuable toxicology studies lack carcinogenicity toxicity knowledge, introduces first The offers alternative dose-dependent carcinogen prediction. Ongoing efforts involve implementing predict expanding application include multiple PFAS co-occurring chemicals.