Yin Liangwei, Traversa Pietro, Elati Mohamed, Moreno Yamir, Marek-Trzonkowska Natalia, Battail Christophe
Université Grenoble Alpes, IRIG, Laboratoire Biosciences et Bioingénierie pour la Santé, UA 13 INSERM-CEA-UGA, 38000 Grenoble, France.
Institute for Biocomputation and Physics of Complex Systems (BIFI), University of Zaragoza, 50018 Zaragoza, Spain.
iScience. 2025 Jul 5;28(8):113061. doi: 10.1016/j.isci.2025.113061. eCollection 2025 Aug 15.
Immunotherapies have recently emerged as a standard of care for advanced cancers, offering remarkable improvements in patient prognosis. However, only a small subset of patients benefit, and robust molecular predictors remain elusive. We present a computational framework leveraging sample-specific gene co-expression networks to identify features predictive of immunotherapy response in kidney cancer. Our results reveal that patients with similar clinical outcomes exhibit comparable gene co-expression patterns. Notably, increased gene connectivity and stronger negative gene-gene associations are hallmarks of poor responders. We further developed sample-specific pathway-level network scores to detect dysregulated biological pathways linked to treatment outcomes. Finally, incorporating these sample-level network features improves the predictive performance of gene expression-based machine learning models. This work highlights the value of personalized gene network features for stratifying patients with cancer and optimizing immunotherapy strategies.
免疫疗法最近已成为晚期癌症的护理标准，显著改善了患者的预后。然而，只有一小部分患者受益，强大的分子预测指标仍然难以捉摸。我们提出了一个计算框架，利用样本特异性基因共表达网络来识别预测肾癌免疫治疗反应的特征。我们的结果表明，具有相似临床结果的患者表现出可比的基因共表达模式。值得注意的是，基因连通性增加和更强的负基因-基因关联是反应不佳者的标志。我们进一步开发了样本特异性通路水平网络评分，以检测与治疗结果相关的失调生物通路。最后，纳入这些样本水平的网络特征提高了基于基因表达的机器学习模型的预测性能。这项工作突出了个性化基因网络特征在对癌症患者进行分层和优化免疫治疗策略方面的价值。
