目的 探讨乌梅治疗溃疡性结肠炎的潜在作用机制，筛选乌梅治疗溃疡性结肠炎的关键作用靶基因，并评估其辅助诊断价值。方法 以GSE87466为发现集，进行差异表达分析，筛选差异表达基因（DEGs）。整合乌梅潜在作用靶点、溃疡性结肠炎疾病靶点及GSE87466-DEGs，获得候选交集基因，并进行基因本体（GO）与京都基因与基因组百科全书（KEGG）富集分析和蛋白质-蛋白质相互作用（PPI）网络构建。基于90个候选交集基因，在GSE87466中采用最小绝对收缩和选择算子（LASSO）、随机森林（RF）、支持向量机（SVM）、XGBoost、决策树（DT）、K近邻（KNN）、神经网络（NNET）、梯度提升机（GBM）、广义线性模型（GLM）及C5.0共10种机器学习算法筛选候选关键靶基因。随后结合GSE87466、GSE38713和GSE75214的表达验证及辅助诊断模型比较，确定最终hub基因。进一步在GSE87466中进行免疫浸润分析及多维加权基因共表达网络分析（WGCNA），以探讨关键基因与疾病表型及免疫炎症微环境之间的关系，并通过分子对接评价乌梅活性成分与候选靶蛋白的潜在结合能力。结果 在GSE87466中共筛得972个DEGs，其中上调620个，下调352个。乌梅潜在作用靶点、溃疡性结肠炎疾病靶点与GSE87466-DEGs取交集后获得90个候选基因。PPI网络包含68个节点和320条边，共筛得36个核心基因。机器学习筛选获得6个候选关键靶基因：编码蛋白溶质载体家族16成员1（SLC16A1）、血管细胞黏附分子1（VCAM1）、水通道蛋白8（AQP8）和双氧化酶2（DUOX2）、ATP结合盒亚家族G成员2（ABCG2）和核受体亚家族1H组成员4（NR1H4）。表达验证显示，SLC16A1和AQP8在溃疡性结肠炎组中低表达，VCAM1和DUOX2在溃疡性结肠炎组中高表达，且在3个数据集中变化趋势一致。模型比较后，最终确定SLC16A1、AQP8、VCAM1和DUOX2为最终hub基因。4基因模型在GSE87466、GSE38713和GSE75214中的曲线下面积（AUC）分别为1.000、0.933、0.988，提示其具有较好的辅助诊断效能。ssGSEA免疫浸润分析显示，GSE87466中27类免疫细胞有26类存在显著差异，且AQP8和SLC16A1与大多数免疫细胞浸润总体呈负相关，而VCAM1和DUOX2总体呈正相关。多维WGCNA显示，MEdarkslateblue模块与疾病状态及中性粒细胞、单核细胞、M1型巨噬细胞、活化树突状细胞和CD4⁺ T细胞呈显著正相关，而MEbrown2模块与上述性状呈显著负相关；VCAM1和DUOX2位于MEdarkslateblue模块，SLC16A1位于MEbrown2模块，AQP8归属于MEdarkslateblue模块但总体呈负向变化趋势。结论 乌梅可能通过调控免疫炎症相关网络干预溃疡性结肠炎的发生发展。SLC16A1、AQP8、VCAM1和DUOX2可能是乌梅干预溃疡性结肠炎的重要关键靶基因，并具有一定的辅助诊断潜力。

Objective To explore the potential mechanism of Fructus Mume in treatment of ulcerative colitis, screen key target genes of Fructus Mume against ulcerative colitis, and evaluate their auxiliary diagnostic value.Methods The dataset GSE87466 was used as the discovery set for differential expression analysis to identify differentially expressed genes (DEGs). Candidate intersecting genes were obtained by integrating potential targets of Fructus Mume, ulcerative colitis-related disease targets and GSE87466-derived DEGs. Gene ontology (GO), Kyoto encyclopedia of genes and genomes (KEGG) enrichment analyses and protein-protein interaction (PPI) network construction were performed on these intersecting genes. Based on the 90 candidate intersecting genes, ten machine learning algorithms including least absolute shrinkage and selection operator (LASSO), random forest (RF), support vector machine (SVM), extreme gradient boosting (XGBoost), decision tree (DT), k-nearest neighbor (KNN), neural network (NNET), generalized boosted regression model (GBM), generalized linear model (GLM) and C5.0 were applied in GSE87466 to select candidate key target genes. Subsequent expression validation across GSE87466, GSE38713 and GSE75214 as well as comparison of auxiliary diagnostic models were conducted to determine the final hub genes. Immune infiltration analysis and multi-dimensional weighted gene co-expression network analysis (WGCNA) were further performed in GSE87466 to investigate the associations between key genes, disease phenotypes and the immune-inflammatory microenvironment. Molecular docking was adopted to assess the potential binding affinity between active ingredients of Fructus Mume and candidate target proteins. Results A total of 972 DEGs were identified in GSE87466, consisting of 620 upregulated and 352 downregulated genes. Ninety candidate genes were obtained after intersecting potential targets of Fructus Mume, ulcerative colitis disease targets and GSE87466 DEGs. The PPI network contained 68 nodes and 320 edges, from which 36 core genes were extracted. Six candidate key target genes were screened via machine learning: Solute carrier family 16 member 1 (SLC16A1), Aquaporin 8 (AQP8), ATP-binding cassette subfamily G member 2 (ABCG2), vascular cell adhesion molecule 1 (VCAM1), nuclear receptor subfamily 1 group H member 4 (NR1H4) and dual oxidase 2 (DUOX2). Expression validation demonstrated that SLC16A1 and AQP8 were downregulated, whereas VCAM1 and DUOX2 were upregulated in UC samples, with consistent expression trends across the three datasets. After model comparison, SLC16A1, AQP8, VCAM1 and DUOX2 were identified as the final hub genes. The four-gene model yielded AUC values of 1.000, 0.933 and 0.988 in GSE87466, GSE38713 and GSE75214 respectively, indicating favorable auxiliary diagnostic performance. Single-sample gene set enrichment analysis (ssGSEA) for immune infiltration revealed significant differences in 26 out of 27 immune cell subtypes in GSE87466. AQP8 and SLC16A1 were generally negatively correlated with the infiltration of most immune cells, while VCAM1 and DUOX2 showed overall positive correlations. Multi-dimensional WGCNA showed that the MEdarkslateblue module was significantly positively correlated with disease status as well as neutrophils, monocytes, M1 macrophages, activated dendritic cells and CD4⁺ T cells, whereas the MEbrown2 module exhibited significant negative correlations with these traits. VCAM1 and DUOX2 belonged to the MEdarkslateblue module, SLC16A1 was assigned to the MEbrown2 module, and AQP8 was classified into the MEdarkslateblue module with an overall negative expression trend. Conclusion Fructus Mume may intervene in the initiation and progression of UC by regulating immune-inflammatory networks. SLC16A1, AQP8, VCAM1 and DUOX2 serve as vital target genes of Fructus Mume for UC intervention and possess potential for auxiliary diagnosis.

R285.5

陕西省科技厅项目（2022JM-506）；陕西省中医药科技创新提质扩能计划项目（TZKN-CXPT-03）