Machine learning-guided discovery of natural anti-inflammatory agents targeting NF-κB p65

Natural products with anti-inflammatory properties have broad applications in biotechnology and healthcare, including nutraceuticals and therapeutics. Here, we present a machine learning-guided experimental framework to identify natural anti-inflammatory agents targeting NF-κB p65. A large-scale dataset of 179,771 protein-compound IC 50 values was compiled from the Anti-Inflammatory Compounds Database and BindingDB entries associated with inflammation-related pathways. Concatenated features of natural products and protein sequences were used to train multilayer perceptron, random forest, and XGBoost regression models. Among these, XGBoost achieved the best predictive performance, with a Pearson correlation coefficient of 0.893, and was applied to predict inhibitory activity of 101 natural products against NF-κB p65, a central transcription factor in inflammatory signaling. Luteolin was predicted to show the strongest inhibitory effect after JSH-23, which served as a positive control. Experimental validation confirmed that luteolin suppressed poly(I:C)-induced p65 activation and reduced expression of NF-κB target genes IL6 , TNF , IL1B , and CXCL8 in human keratinocytes and three-dimensional human epidermal models. Structural analysis using AlphaFold 3 supported luteolin binding to p65 and interference with DNA interaction. Together, these results demonstrate the utility of integrating machine learning with biological assays to accelerate discovery of natural anti-inflammatory agents for biotechnological applications.