Exploring the Molecular Mechanisms of Weifuchun in the Treatment of Chronic Atrophic Gastritis Based on Single-cell Sequencing and Machine Learning.

ETHNOPHARMACOLOGICAL RELEVANCE: Chronic atrophic gastritis (CAG) is a common precancerous lesion of gastric cancer, for which there is currently a lack of effective drugs to reverse its pathological progression. Weifuchun (WFC), as a classical traditional Chinese medicine (TCM) compound, has demonstrated definitive clinical efficacy in treating CAG, although its molecular mechanisms remain unclear. AIM OF THE STUDY: To explore the potential molecular mechanisms of WFC in the treatment of CAG. MATERIALS AND METHODS: This study conducted single-cell RNA sequencing (scRNA-seq) on gastric mucosal tissues from patients with chronic superficial gastritis (CSG) and CAG, mapping gastric mucosal cell populations and screening differentially expressed genes. The CAG transcriptome data from six integrated databases from Gene Expression Omnibus (GEO) were combined with the databases from Traditional Chinese Medicine Systems Pharmacology Database and Analysis Platform (TCMSP) and relevant literature to identify active components and potential targets of WFC, constructing a "active components-cell populations-intersected target genes" regulatory network. Eight machine learning models (including random forest) were employed to predict the efficacy of key target genes. Finally, a cell model was established using MNNG-induced GES-1 cells, with potential target protein expression validated by Western blot. RESULTS: Single- cell sequencing identified 17 cell types, with epithelial cells constituting the highest proportion (86.96%). In the CAG group, T/NK cells, plasma cells, and fibroblasts were increased as abserved, while goblet cells and enterocytes appeared exclusively in CAG samples. Pseudotime analysis revealed a preferential polarization of mucosal-associated invariant T (MAIT) cells toward a Th17- like phenotype in CAG, accompanied by a decline in IFNG+ effector CD8+ T cells. Cellular interaction analysis highlighted fibroblasts, macrophages, and dendritic cells as central hubs, with key ligand- receptor pairs (e.g., CCL5/CCR5, ADM/RAMP3) implicated in pro-inflammatory crosstalk. Thirteen active components of WFC were identified, and their target genes overlapped with CAG differentially expressed genes from single- cell sequencing, yielding five common targets (PDE4D, JUN, KCNH2, NFKBIA, MMP12). Further intersection with GEO- derived CAG DEGs identified JUN and MMP12 as potential targets. Machine learning demonstrated that the random forest model exhibited optimal classification performance (AUC = 1.000), with JUN being the candidate feature (single- gene discriminant accuracy approaching 95%). JUN, NFKBIA, and MMP12 formed the potential feature set. In vitro experiments showed that WFC concentration-dependently inhibited the proliferation of MNNG-induced GES-1 cells and significantly downregulated MMP12 protein expression (P < 0.05), while no significant effects were observed on JUN, NFKBIA, PDE4D, or KCNH2 protein levels. CONCLUSION: This study revealed the cellular atlas, immune microenvironment shifts, and potential target genes of WFC in treating CAG, demonstrating that WFC may delay CAG progression by downregulating MMP12 expression. Machine learning identified JUN as the candidate predictive gene. This research provides a preliminary approach for precision mechanism studies of traditional Chinese medicine formulations, encompassing "single- cell atlas - multi-source data integration - machine learning prediction - experimental validation".