AHR/NRF2 Dual Agonist Prediction and Natural Compound Screening Based on Machine Learning: A New Strategy for the...

In the treatment of atopic dermatitis (AD), synergistic activation of the aryl hydrocarbon receptor (AHR)/nuclear factor erythroid 2-related factor 2 (NRF2) pathways represents a promising strategy. However, known dual agonists are limited, and traditional screening methods are inefficient. Therefore, this study developed machine learning models to predict AHR/NRF2 dual agonists using molecular descriptors and fingerprints. All models achieved area under the receiver operating characteristic curve (AUC) values above 0.86, indicating good classification performance. The optimal AHR model showed an accuracy (ACC) of 0.811 and an AUC of 0.878, while the best NRF2 model yielded an ACC of 0.839 and an AUC of 0.907. Based on this model, compounds with a low fraction of sp³-hybridized carbons, moderate hydrophobicity, limited alkyl chains, and highly conjugated structures tend to act as AHR/NRF2 dual agonists. Finally, this study screened 1011 potential natural AHR/NRF2 dual agonists suitable for drug development. Among these, 2-arylbenzofurans, alkaloids, phenanthrenes, flavones, and furocoumarins demonstrated particular advantages. For validation, Indirubin, imperatorin and 3'-O-Methylbutastatin III were first discovered as AHR/NRF2 dual agonists in HaCaT cells. This work provides a robust predictive tool, clarifies key molecular features of dual agonists, and may support the discovery of anti-AD agents.