Fehse Leon, Tajabadi Mohammad, Martin Roman, Holzmann Hajo, Heider Dominik
University of Münster, Institute of Medical Informatics, Albert-Schweitzer-Campus 1, Münster, 48149, North Rhine-Westphalia, Germany.
Institute for Computer Science, Heinrich-Heine-University Düsseldorf, Graf-Adolf-Str. 63, Düsseldorf, 40215, North Rhine-Westphalia, Germany.
Comput Struct Biotechnol J. 2025 Jul 31;27:3456-3463. doi: 10.1016/j.csbj.2025.07.031. eCollection 2025.
Count data, such as gene expression and microbiome composition, play a significant role in various diseases, including cancer, obesity, inflammatory bowel disease, and mental health disorders. For instance, understanding the differences in microbial abundance between patients is essential for uncovering the microbiome's impact on these conditions. Differential abundance analysis (DAA) can detect significant changes between groups of patients. However, since individuals have unique microbial fingerprints that could potentially be identifiable, microbiome data must be treated as sensitive patient data, which poses problems for collaborative studies in the medical field. In this work, we introduce gLinDA, a global differential abundance analysis tool that employs a privacy-preserving swarm learning approach for the analysis of distributed datasets. gLinDA maintains predictive performance while safeguarding patient sensitive data.
计数数据，如基因表达和微生物组组成，在包括癌症、肥胖症、炎症性肠病和精神健康障碍在内的各种疾病中发挥着重要作用。例如，了解患者之间微生物丰度的差异对于揭示微生物组对这些病症的影响至关重要。差异丰度分析（DAA）可以检测患者组之间的显著变化。然而，由于个体具有独特的微生物指纹，可能具有可识别性，因此微生物组数据必须被视为敏感的患者数据，这给医学领域的合作研究带来了问题。在这项工作中，我们引入了gLinDA，这是一种全局差异丰度分析工具，它采用隐私保护群体学习方法来分析分布式数据集。gLinDA在保护患者敏感数据的同时保持预测性能。
