文献速递类风湿关节炎

利用全血转录组学数据的机器学习分析识别接受阿达木单抗治疗的类风湿关节炎患者疗效预测生物标志物

Identifying Predictive Biomarkers of Response in Patients With Rheumatoid Arthritis Treated With Adalimumab Using Machine Learning Analysis of Whole-Blood Transcriptomics Data.
PubMed 文献 · Arthritis Rheumatol · 2025-08-04

本研究纳入100例起始TNFi治疗的类风湿关节炎患者,在基线及3个月随访时进行RNA测序,6个月评估疗效,构建机器学习分类器识别疗效预测生物标志物。97例质控合格样本的差异基因分析识别出84个与治疗反应相关的基因,随机森林分类器曲线下面积最高达0.86;网络分析进一步识别出MZB1为机器学习未捕获的新生物标志物,其与B细胞发育和抗体产生相关,并与抗药抗体形成相关、影响疗效。

要点

  • 97例质控合格样本的差异基因分析识别出84个与治疗反应相关的基因。
  • 随机森林分类器预测准确性高,受试者工作特征曲线下面积最高达0.86。
  • 网络分析识别出MZB1为机器学习未捕获的新生物标志物,与抗药抗体形成相关并影响疗效。

核心结论

  • 本研究增进了对类风湿关节炎治疗中转录组改变的理解,并加深了对治疗反应机制的认识。
  • 尽管所识别的基因特征需要独立重复验证,本研究为起始TNFi治疗的类风湿关节炎患者个体化治疗策略铺路提供了起点。

研究设计与样本

  • 队列纳入100例起始TNFi治疗的类风湿关节炎患者,在6个月时评估治疗反应。
  • 对基线(治疗前)和3个月随访样本进行RNA测序。
  • 构建机器学习分类器以识别治疗结局的预测生物标志物,并对生物标志物进行网络分析以阐明最具影响力的生物标志物,随后通过生存分析确认。

主要发现

  • 在97例通过质控的样本中进行差异基因表达分析,识别出84个与治疗反应相关的基因。
  • 随机森林分类器预测准确性高,受试者工作特征曲线下面积最高达0.86,并识别出对治疗结局有贡献的基因。
  • 网络分析进一步阐明了基因相互作用,突出显示边缘区B细胞和B1细胞特异性蛋白1(MZB1)为机器学习单独分析未捕获的新生物标志物。
  • MZB1在B细胞发育和抗体产生中的作用与抗药抗体形成相关,并影响治疗疗效。

临床怎么用

  • 原文未据此给出推荐。

局限性

  • 所识别的基因特征需要独立重复验证。

以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。

原文

OBJECTIVE: Tumornecrosis factor inhibitors (TNFi) have significantly improved rheumatoid arthritis (RA) management, yet variability in patient response remains a substantial challenge, with approximately 40% of patients discontinuing TNFi due to nonresponse or adverse effects. This study aimed to identify biomarkers predictive of adalimumab treatment response using whole-blood transcriptomics, leveraging machine learning models for data mining observed by targeted statistical analysis. METHODS: A cohort of patients with RA starting TNFi therapy (n = 100) was assessed for treatment response at 6 months, with RNA sequencing performed on baseline (pretreatment) and 3-month follow-up samples. Machine learning classifiers were built to identify predictive biomarkers for treatment outcomes. This was observed by a network analysis on the biomarkers to elucidate the most influential biomarker, which was subsequently confirmed through survival analysis. RESULTS: Differential gene expression analysis in 97 samples passing quality control identified 84 genes associated with treatment response. Random forest classifiers achieved high predictive accuracy with area under the receiver operating characteristic curves up to 0.86, identifying genes contributing to treatment outcomes. Network analysis further elucidated gene interactions, highlighting marginal zone B And B1 cell-specific protein 1 (MZB1) as a novel biomarker not captured by machine learning alone. MZB1's role in B cell development and antibody production was associated with antidrug antibody formation, impacting treatment efficacy. CONCLUSION: This study advances the understanding of transcriptomic alterations in RA treatment and enhances our understanding of treatment response mechanisms. Although the gene signatures identified require independent replication, the study serves as a starting point to pave the way for personalized therapeutic strategies in patients commencing TNFi therapy in RA.

来源:Arthritis Rheumatol,查看原文。版权归原作者所有。

引用

  1. Yap CF, Nair N, Morgan AW, et al. Identifying Predictive Biomarkers of Response in Patients With Rheumatoid Arthritis Treated With Adalimumab Using Machine Learning Analysis of Whole-Blood Transcriptomics Data. Arthritis Rheumatol. 2025. PMID: 40415615. doi:10.1002/art.43255

供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。