Hong Kong Baptist University

Long-term exposure to fine particulate matter (PM2.5) is associated with respiratory and cardiovascular diseases. PM2.5 consists of a complex mixture of organic and inorganic species, with toxicity varying based on its chemical composition, sources, and physicochemical properties. This study investigates the oxidative potential (OP), cellular oxidative stress, and inflammatory response induced by five distinct chemical fractions of urban PM2.5: water-soluble total, water-soluble metals, water-soluble non-metal, lipid-soluble, and total PM2.5 extract. We also analyzed the synergistic and antagonistic interactions among these fractions contributing to overall PM2.5 toxicity. Comprehensive chemical characterization and OP analysis of PM2.5 extracts revealed that metals primarily drive dithiothreitol (DTT) consumption, while organics predominantly contribute to hydroxyl radical (∙OH) generation. Notably, high PM2.5 samples exhibited significant antagonistic interactions between water-soluble metals and organic fractions in the generation of ∙OH. The water-soluble total fraction induced the highest levels of TNF-α secretion and upregulated the expression of genes associated with inflammation and oxidative stress, including Cxcl2, Hmox-1, and Cyp1a1, emphasizing its dominant role in PM2.5-induced cytotoxicity. Synergistic upregulation of Hmox-1 expression was observed between water-soluble metals and non-metal fractions, whereas Cxcl2 expression was antagonistically modulated. Conversely, the lipid-soluble fraction exhibited an antagonistic effect on TNF-α secretion and oxidative stress gene expression relative to the water-soluble total fraction. These findings highlight the pivotal role of water-soluble components in PM2.5 toxicity and provide a comprehensive framework for understanding the individual and combined effects of chemical fractions on PM2.5-induced toxicity, which is vital for accurately assessing its impact on human health.

Accurate classification of encrypted IoT traffic remains challenging due to evolving applications and distribution shifts. This work presents a self-updating hybrid meta-learning framework that integrates Bayesian Neural Networks (BNN) for uncertainty-aware update triggering with a Random Forest meta-classifier for robust decision fusion. The proposed design improves scalability and interpretability through feature-importance analysis and lightweight ensemble learning. Prediction instability is quantified using the Hellinger distance, avoiding normalization overhead and enabling an adaptive familiarity score via a tunable parameter α. Experimental results on encrypted traffic datasets demonstrate significant gains in reliability, achieving up to 95.7% accuracy and 0.95 macro-F1, and effective selective retraining under distribution shifts.