Institute of Evolutionary Biology (UPF-CSIC), Department of Experimental and Health Sciences, Pompeu Fabra University, Barcelona Biomedical Research Park, Doctor Aigüader 88, 08003 Barcelona, Spain
Biol Open. 2020 Oct 19;9(10):bio056176. doi: 10.1242/bio.056176.
The study of morphological modularity using anatomical networks is growing in recent years. A common strategy to find the best network partition uses community detection algorithms that optimize the modularity Q function. Because anatomical networks and their modules tend to be small, this strategy often produces two problems. One is that some algorithms find inexplicable different modules when one inputs slightly different networks. The other is that algorithms find asymmetric modules in otherwise symmetric networks. These problems have discouraged researchers to use anatomical network analysis and boost criticisms to this methodology. Here, I propose a node-based informed modularity strategy (NIMS) to identify modules in anatomical networks that bypass resolution and sensitivity limitations by using a bottom-up approach. Starting with the local modularity around every individual node, NIMS returns the modular organization of the network by merging non-redundant modules and assessing their intersection statistically using combinatorial theory. Instead of acting as a black box, NIMS allows researchers to make informed decisions about whether to merge non-redundant modules. NIMS returns network modules that are robust to minor variation and does not require optimization of a global modularity function. NIMS may prove useful to identify modules also in small ecological and social networks.
近年来，使用解剖学网络研究形态模块化的方法正在不断发展。一种常见的找到最佳网络划分的策略是使用社区检测算法来优化模块性 Q 函数。由于解剖学网络及其模块往往很小，因此该策略经常产生两个问题。一个问题是，当输入稍微不同的网络时，一些算法会找到无法解释的不同模块。另一个问题是，算法会在原本对称的网络中找到不对称的模块。这些问题阻碍了研究人员使用解剖学网络分析，并加剧了对这种方法的批评。在这里，我提出了一种基于节点的信息模块化策略（NIMS），该策略通过使用自下而上的方法来绕过分辨率和灵敏度限制，从而识别解剖学网络中的模块。从每个单独节点的局部模块性开始，NIMS 通过合并非冗余模块并使用组合理论统计评估它们的交集来返回网络的模块化组织。与作为黑盒的算法不同，NIMS 允许研究人员就是否合并非冗余模块做出明智的决策。NIMS 返回的网络模块对微小变化具有鲁棒性，并且不需要优化全局模块性函数。NIMS 可能有助于识别小型生态和社交网络中的模块。
