Table1_Integration of three machine learning algorithms identifies characteristic RNA binding proteins linked with diagnosis, immunity and pyroptosis of IgA nephropathy.XLSX

Objective: RNA-binding proteins (RBPs) are essential for most post-transcriptional regulatory events, which exert critical roles in nearly all aspects of cell biology. Here, characteristic RBPs of IgA nephropathy were determined with multiple machine learning algorithms.
Methods: Our study included three gene expression datasets of IgA nephropathy (GSE37460, GSE73953, GSE93798). Differential expression of RBPs between IgA nephropathy and normal samples was analyzed via limma, and hub RBPs were determined through MCODE. Afterwards, three machine learning algorithms (LASSO, SVM-RFE, random forest) were integrated to determine characteristic RBPs, which were verified in the Nephroseq database. Immune cell infiltrations were estimated through CIBERSORT. Utilizing ConsensusClusterPlus, IgA nephropathy were classified based on hub RBPs. The potential upstream miRNAs were predicted.
Results: Among 388 RBPs with differential expression, 43 hub RBPs were determined. After integration of three machine learning algorithms, three characteristic RBPs were finally identified (DDX27, RCL1, and TFB2M). All of them were down-regulated in IgA nephropathy than normal specimens, with the excellent diagnostic efficacy. Additionally, they were significantly linked to immune cell infiltrations, immune checkpoints, and pyroptosis-relevant genes. Based on hub RBPs, IgA nephropathy was stably classified as two subtypes (cluster 1 and 2). Cluster 1 exhibited the relatively high expression of pyroptosis-relevant genes and characteristic RBPs. MiR-501-3p, miR-760, miR-502-3p, miR-1224-5p, and miR-107 were potential upstream miRNAs of hub RBPs.
Conclusion: Collectively, our findings determine three characteristic RBPs in IgA nephropathy and two RBPs-based subtypes, and thus provide a certain basis for further research on the diagnosis and pathogenesis of IgA nephropathy.

研究目的：RNA结合蛋白（RNA-binding proteins, RBPs）是绝大多数转录后调控事件的核心参与者，在细胞生物学的几乎所有环节中发挥关键作用。本研究通过多种机器学习算法，明确了IgA肾病（IgA nephropathy）的特征性RNA结合蛋白。
研究方法：本研究纳入了3组IgA肾病的基因表达数据集（GSE37460、GSE73953、GSE93798）。通过limma分析IgA肾病样本与正常对照样本中RBPs的差异表达水平，并通过MCODE算法筛选得到核心RNA结合蛋白（hub RBPs）。随后整合LASSO、SVM-RFE、随机森林（random forest）三种机器学习算法，确定特征性RBPs，并在Nephroseq数据库中对其进行验证。通过CIBERSORT算法评估免疫细胞浸润情况，利用ConsensusClusterPlus工具基于核心RBPs对IgA肾病样本进行分型，并预测其潜在上游微小RNA（microRNAs, miRNAs）。
研究结果：在388个差异表达的RBPs中，共筛选得到43个核心RBPs。通过整合三种机器学习算法，最终确定3个特征性RBPs：DDX27、RCL1及TFB2M。相较于正常标本，这3个基因在IgA肾病样本中均呈低表达，且具备优异的诊断效能。此外，它们与免疫细胞浸润、免疫检查点及细胞焦亡（pyroptosis）相关基因显著相关。基于核心RBPs，IgA肾病可被稳定分为两个亚型（cluster 1与cluster 2）；其中cluster 1的细胞焦亡相关基因及特征性RBPs表达水平相对更高。最终预测得到miR-501-3p、miR-760、miR-502-3p、miR-1224-5p及miR-107为核心RBPs的潜在上游微小RNA。
研究结论：综上，本研究明确了IgA肾病中的3个特征性RBPs及2个基于RBPs的分型亚型，为IgA肾病的诊断及发病机制的后续研究提供了一定理论依据。