Analysis of interactions of particle-associated oxidative potential sources using multilayer perceptron neural networks: A case study in Shenyang, China

The oxidative potential (OP) of particulate matter (PM) is a possible indicator for assessing the oxidative-imbalance risk caused by PM exposure. The OP contributions of different PM sources exhibit nonlinear relationships, and the specific patterns and intensities of these interactions remain unclear. This study sampled total suspended particulates (TSPs) seasonally in 2015 in Shenyang, a major industrial city in China. Chemical analyses were performed on samples, and six potential sources were identified via positive matrix factorization: automobile exhaust and road dust, biomass burning, secondary pollution, coal combustion, diesel combustion, and soil. The OPs of TSP samples were quantified using volume-based dithiothreitol assay. A multilayer perceptron, an artificial neural network, was used to model relationships among the sources and OP_DTTv (the sampling volume as a proxy for the OP level) considering nonlinear interactions between sources. The trained model was used to analyze potential pairwise interactions, whose strengths were determined by calculating interaction factors. Simulation of a typical winter reveals significant synergistic effects and weak antagonistic effects between certain source combinations, while simulation of a typical summer shows weak synergistic and antagonistic effects. Real-world sampling results confirm that some source concentration combinations are consistent with the simulated interactions. This study highlights interactions between source contributions to OP, identifies combinations with notable synergistic or antagonistic effects, and emphasizes the importance of comprehensive source control strategies for mitigating risks associated with synergistic effects.