Singlan Nina, Abou Choucha Fadi, Pasquier Claude
Université Côte d'Azur, CNRS, i3S, 06560, Valbonne, France.
Sci Rep. 2025 Apr 2;15(1):11360. doi: 10.1038/s41598-025-95749-6.
Real-world networks, such as biological networks, often exhibit complex structures and have attributes associated with nodes, which leads to significant challenges for analysis and modeling. Community detection algorithms can help identify groups of nodes of particular importance. However, traditional methods focus primarily on topological information, overlooking the importance of attribute-based similarities. This limitation hinders their ability to identify functionally coherent subnetworks. To address this, we propose a new scoring method for graph partitioning on the basis of a novel similarity function between node attributes. We then adapt the Louvain algorithm to optimize this scoring function, enabling the identification of communities that are both densely connected and functionally coherent. Extensive experiments on diverse biological networks, including artificial and real-world datasets, demonstrate the superiority of our approach over state-of-the-art methods. By leveraging both topological and attribute-based information, our approach provides a powerful tool for uncovering biologically meaningful modules and gaining deeper insights into complex biological processes.
现实世界中的网络，如生物网络，通常呈现出复杂的结构，并且具有与节点相关联的属性，这给分析和建模带来了重大挑战。社区检测算法有助于识别特别重要的节点组。然而，传统方法主要关注拓扑信息，而忽略了基于属性的相似性的重要性。这种局限性阻碍了它们识别功能连贯的子网的能力。为了解决这个问题，我们基于一种新的节点属性之间的相似性函数，提出了一种用于图划分的新评分方法。然后，我们对Louvain算法进行了调整，以优化这个评分函数，从而能够识别出既紧密连接又功能连贯的社区。在包括人工和真实世界数据集在内的各种生物网络上进行的大量实验表明，我们的方法优于现有方法。通过利用拓扑和基于属性的信息，我们的方法为揭示具有生物学意义的模块和深入了解复杂的生物过程提供了一个强大的工具。
