Weighted gene co-expression network analysis and machine learning identification of key genes in rheumatoid arthritis synovium

doi:10.12307/2025.244

Abstract

Abstract: BACKGROUND: Rheumatoid arthritis is a condition that affects the entire immune system in the body and is known for causing inflammatory hyperplasia in the joints and destruction of articular cartilage. The pathogenesis of rheumatoid arthritis is still unclear; therefore, there is an urgent need to discover new highly sensitive and specific diagnostic biomarkers.
OBJECTIVE: To identify and screen key genes in the synovium of rheumatoid arthritis patients using bioinformatics techniques and machine learning algorithms and to construct and validate a rheumatoid arthritis prediction model.
METHODS: Three datasets containing synovial tissue samples from rheumatoid arthritis patients (GSE77298, GSE55235, GSE55457) were downloaded from the Gene Expression Omnibus (GEO) database. GSE77298 and GSE55235 were used as the training set, while GSE55457 served as the test set, with a total of 66 samples, including 39 samples from rheumatoid arthritis patients and 27 normal synovial samples. Differentially expressed genes in the training set were selected using R language, and then the weighted gene co-expression network analysis was used to modularize the genes in the training set. The most relevant module was selected, and feature genes within this module were identified. Differentially expressed genes and the feature genes from the module were intersected for the subsequent machine learning analysis. Three machine learning methods, namely the least absolute shrinkage and selection operator algorithm, support vector machine with recursive feature elimination, and random forest algorithm, were employed to further analyze the intersected genes and identify the hub genes. The hub genes obtained from these three machine learning algorithms were intersected again to obtain the key genes in the synovium of rheumatoid arthritis. A predictive rheumatoid arthritis model was constructed using these key genes as variables, and the risk of developing rheumatoid arthritis in patients was inferred based on the model. The receiver operating characteristic curve was used to determine the diagnostic value of the rheumatoid arthritis prediction model and its key genes.
RESULTS AND CONCLUSION: Through the differential analysis, a total of 730 differentially expressed genes were identified in the training set, and 185 feature genes were identified in the weighted gene co-expression network analysis feature modules. There were 159 intersected genes obtained. There were 4 hub genes identified by the least absolute shrinkage and selection operator algorithm, 11 hub genes by the support vector machine with recursive feature elimination algorithm, and 5 hub genes by the random forest algorithm. After intersection, 2 key genes (TNS3 and SDC1) were obtained. Based on the two key genes, a nomogram model was constructed in the training and test sets, with good fit between the calibration prediction curve and the standard curve, and good clinical efficacy in predicting the onset of rheumatoid arthritis. These findings indicate that TNS3 and SDC1, obtained based on bioinformatics and machine learning algorithms, may become key targets for the diagnosis and treatment of rheumatoid arthritis.

Key words: weighted gene co-expression network, machine learning algorithm, rheumatoid arthritis, key gene, prediction model

CLC Number:

Wu Yingkai, Shi Gaolong, Xie Zonggang . Weighted gene co-expression network analysis and machine learning identification of key genes in rheumatoid arthritis synovium[J]. Chinese Journal of Tissue Engineering Research, 2025, 29(2): 294-301.

Figures/Tables 7

2.1 差异表达基因筛选根据筛选标准，将类风湿关节炎滑膜样本同正常滑膜样本中的基因表达量进行差异分析，共获得730个差异表达基因，其中上调451个，下调279个。差异表达基因分层聚类图及火山图(图1)。 2.2 差异表达基因功能富集分析基因本体论分析结果(图2)表明，差异表达基因主要参与白细胞迁移、免疫反应调节信号通路、免疫调节细胞表面受体信号通路、免疫反应激活信号通路、免疫反应激活细胞表面受体信号通路等生物学过程；主要定位于肌原纤维节、肌原纤维、浆细胞外膜、肌动蛋白)、胶原蛋白的细胞外基质等细胞成分；主要参与免疫受体活性、肝素结合、葡萄聚糖结合、G蛋白受体结合、免疫因子结合、免疫因子活动、肌动蛋白结合等分子功能。KEGG分析(图3)显示，差异表达基因主要富集于成骨细胞分化、利什曼病、B细胞受体通路、病毒蛋白与细胞因子及细胞因子受体的相互作用、Fc γ受体介导的吞噬、趋化因子通路、扩张型心肌病、白细胞经内皮细胞迁移、自然杀伤细胞调节的炎症反应、炎症因子与炎症因子受体相互作用等相关信号通路。 2.3 WGCNA分析及交集基因获取对GSE77298和GSE55235合并后表达矩阵进行数据预处理，取绝对中位差前75%基因纳入WGCNA分析，使用WGCNA包对预处理后的数据进行聚类。然后进行平均连通性及无标尺度指数分析，当β=5时，网络达到无标尺度拓扑阈值0.9，且网络平均连通性最高。使用动态切割树进行切割和计算，最后获得32个模块，将32个模块同正常组和类风湿关节炎组进行相关性分析，得到基因模块同临床分类相关性热图。选择与类风湿关节炎相关性最强的黑色模块作为关键模块(r=0.83，P < 0.000 1)，以MM(基因与模块相关性) > 0.8作为过滤条件，获得黑色模块中特征基因185个。将差异基因同特征基因取交集，得到159个交集基因用于机器学习，见图4。 2.4 机器学习算法筛选枢纽基因对上述159个基因进行机器学习，在LASSO回归模型中，当选择最小均方误差对数λ时，回归模型共筛选出4个枢纽基因。支持向量机-递归特征消除模型分析发现，最佳点在0.044时5折交叉验证误差最小，共筛选出11个枢纽基因。Randomforest算法共获得5个枢纽基因。将支持向量机-递归特征消除、Randomforest和LASSO分析所筛选的基因取交集，共得到2个关键基因，即TNS3和SDC1，见图5。 2.5 类风湿关节炎预测模型的构建与验证基于TNS3、SDC1两个关键基因，应用rms包在训练集及验证集中构建类风湿关节炎列线图预测模型；并使用Calibration校准曲线预测模型准确性，临床决策曲线分析明确其临床决策效能，结果显示预测模型具有较好的预测准确性及临床决策效能，见图6。 2.6 关键基因和预测模型的诊断效能评价在训练集中TNS3、SDC1的曲线下面积分别为0.965和0.984，预测模型的曲线下面积为0.993；在验证集中TNS3、SDC1的曲线下面积均为0.931，预测模型的曲线下面积为0.946，见图7。 2.7 免疫细胞浸润情况检测通过对类风湿关节炎及正常滑膜组织免疫细胞比例以及差异进行分析，发现与正常组织相比，类风湿关节炎滑膜组织中记忆B细胞、浆细胞、滤泡性T细胞、辅助性T细胞、激活的自然杀伤细胞、单核细胞、M0期巨噬细胞、M2期巨噬细胞、休眠期树突状细胞、休眠期肥大细胞均有显著变化，差异有显著性意义(图8)。"

References

[1] LI Z, CHEN Y, ZULIPIKAER M, et al. Identification of PSMB9 and CXCL13 as Immune-related Diagnostic Markers for Rheumatoid Arthritis by Machine Learning.Curr Pharm Des. 2022;28(34):2842-2854.
[2] JANG S, KWON EJ, LEE JJ. Rheumatoid Arthritis:Pathogenic Roles of Diverse Immune Cells. Int J Mol Sci. 2022;23(2):905.
[3] 宋世雷,陈跃平,陈锋.基于生物信息与机器学习分析类风湿关节炎疾病特征基因与免疫细胞的关系[J].中国免疫学杂志,2023,39(12):2470-2476.
[4] KARAMI J, ASLANI S, JAMSHIDI A, et al. Genetic implications in the pathogenesis of rheumatoid arthritis:an updated review.Gene. 2019;702:8-16.
[5] 钱晓芬,曾平,刘金富,等.差异表达基因联合加权共表达网络分析筛选骨关节炎滑膜中的关键基因 [J].中国组织工程研究,2022,26(33):5342-5349.
[6] 李文静,杨杰,陈冬梅.WGCNA与机器学习算法识别结直肠癌发病机制中潜在PANoptosis关键基因[J].东南大学学报(医学版),2023,42(5):688-696.
[7] GUO S, WU J, ZHOU W, et al.Identification and analysis of key genes associated with acute myocardial infarction by integrated bioinformatics methods.Medicine(Baltimore).2021;100(15):e25553.
[8] AKI T, HASHIMOTO K, OGASAWARA M, et al. A whole-genome transcriptome analysis of articular chondrocytes in secondary osteoarthritis of the hip. PLoS One. 2018; 13(6):e0199734.
[9] LI G, LI X, YANG M, et al. Prediction of biomarkers of oral squamous cell carcinoma using microarray technology. Sci Rep. 2017; 7:42105.
[10] 李慧月,王沐榛,喻叶,等.加权基因共表达网络分析筛选胰腺癌肿瘤免疫相关基因[J].华南师范大学学报(自然科学版), 2021,53(6):61-67.
[11] AUWUL MR, RAHMAN MR, GOV E, et al. Bioinformatics and machine learning approach identifies potential drug targets and pathways in COVID-19. Brief Bioinform. 2021;22(5):bbab120.
[12] 刘孝生,魏东升,何信用,等.基于综合生物信息学和机器学习算法构建衰老相关分泌表型的骨关节炎预测模型[J].中国医科大学学报,2023,53(12):1092-1097.
[13] KANG K, XIE F, MAO J, et al. Significance of tumor mutation burden in immune infiltration and prognosis in cutaneous melanoma. Front Oncol. 2020;10:573141
[14] 范氏绒.昆断益母方调控Treg/Th17细胞失衡和抗类风湿关节炎的机制研究[D].广州:广州中医药大学,2021.
[15] LIN YJ, ANZAGHE M, SCHÜLKE S. Update on the Pathomechanism, Diagnosis, and Treatment Options for Rheumatoid Arthritis.Cells. 2020;9(4):880.
[16] 董文哲,刘健,端淑杰.新风胶囊通过调节SOCS1/STAT3/miR326改善类风湿关节炎患者Th17/Treg失衡状态[J].免疫学杂志,2018,34(6):499-506.
[17] ZHANG W, XIE X, HUANG Z, et al. The integration of single-cell sequencing, TCGA, and GEO data analysis revealed that PRRT3-AS1 is a biomarker and therapeutic target of SKCM. Front Immunol. 2022;13: 919145.
[18] CHOI RY, COYNER AS, KALPATHY-CRAMER J, et al. Introduction to Machine Learning, Neural Networks,and Deep Learning. Transl Vis Sci Technol. 2020;9(2):14.
[19] YANG Y, YI X, CAI Y, et al. Immune-Associated Gene Signatures and Subtypes to Predict the Progression of Atherosclerotic Plaques Based on Machine Learning. Front Pharmacol. 2022;13:865624
[20] 夏天,李炳霖,肖发源,等.类风湿关节炎铁死亡特征基因 CeRNA 网络构建及免疫表现[J].中国组织工程研究,2024, 28(16):2561-2567.
[21] SUN H, PENG G, CHEN K, et al. Identification of EGFR as an essential regulator in chondrocytes ferroptosis of osteoarthritis using bioinformatics,in vivo,and in vitro study. Heliyon. 2023;9(9):e19975.
[22] 史阳.TNS3及其相关超级增强子在LMK-235抑制食管鳞状细胞癌增殖中的机制研究[D].郑州:郑州大学,2021.
[23] WU Y, ZHANG HQ, TANG M, et al. Abnormal TNS3 gene methylation in patients with congenital scoliosis. BMC Musculoskelet Disord. 2022;23(1):797.
[24] TOUAITAHUATA H, MOREL A, URBACH S, et al.Tensin 3 is a new partner of Dock5 that controls osteoclast podosome organization and activity. J Cell Sci. 2016;129(18):3449-3461.
[25] 路帅,李文杰,徐紫薇,等.乳腺癌关键基因SDC1的临床预后意义及其与M1型巨噬细胞相关性分析[J].齐齐哈尔医学院学报,2023,44(11):1001-1010.
[26] 李哲杰.SDC1通过外排EGFR抑制食管癌恶性进展分子机制研究[D].济宁:济宁医学院,2023.
[27] MEYER A, SIENES RE, NIJIM W, et al. Syntenin-1-mediated arthritogenicity is advanced by reprogramming RA metabolic macrophages and Th1 cells. Ann Rheum Dis. 2023;82(4):483-495.
[28] ALZOUBI O, MEYER A, GONZALEZ TP, et al. Significance of IL-34 and SDC-1 in the pathogenesis of RA cells and preclinical models. Clin Immunol. 2023;251:109635.
[29] CHEN D, LIU J, ZANG L, et al. Integrated machine learning and bioinformatic analyses constructed a novel stemness-related classifier to predict prognosis and immunotherapy responses for hepatocellular carcinoma patients. Int J Biol Sci. 2022;18(1):360-373.
[30] LIU Z, MI M, LI X, et al. A lncRNA prognostic signature associated with immune infiltration and tumour mutation burden in breast cancer. J Cell Mol Med. 2020;24(21):12444-12456.