Single-Cell and Machine Learning Analyses Identify a PFKFB3-Centered Regulatory Network and Potential Salidroside Interaction in Coronary Heart Disease

Coronary heart disease (CHD) is a leading cause of morbidity and mortality, driven by metabolic remodeling, vascular inflammation, and perivascular adipose tissue (PVAT) dysfunction. We integrated bulk transcriptomic datasets to develop a machine learning–based diagnostic model, evaluated 113 algorithms, and identified a seven-gene signature (PYGL, PTGS2, PFKFB3, MMP9, CYP1B1, CXCR1, ABCB1) with robust predictive performance. Single-cell RNA sequencing of coronary PVAT revealed substantial cellular heterogeneity and prioritized PFKFB3 as a hub linking glycolytic activity to NF-κB regulon activity. Macrophage-centered communication via SPP1, MIF, and other pathways was enhanced in disease conditions. Virtual knockout of PFKFB3 induced transcriptional changes enriched in immune activation, phagocytosis, and oxidative stress, while molecular dynamics simulations indicated stable salidroside binding within PFKFB3. Together, these analyses provide a multi-layered framework connecting glycolytic remodeling, inflammatory transcriptional activity, and intercellular signaling in CHD. The findings support PFKFB3 as a potential biomarker and mechanistic hub, and suggest that salidroside may modulate its activity. This study offers an integrative computational foundation for future experimental validation and mechanistic exploration of PVAT dysfunction in CHD.