Constructing ecological networks has become an indispensable approach in understanding how different taxa interact. However, the methods used to generate data in network research varies widely among studies, potentially limiting our ability to compare results meaningfully. In particular, methods of classifying nodes vary in their precision, likely altering the architecture of the network studied. For example, rather than being classified as Linnaean species, taxa are regularly assigned to morphospecies in observational studies, or to Molecular Operational Taxonomic Units (MOTUs) in molecular studies, with the latter defined based on an arbitrary threshold of sequence similarity. Although the use of MOTUs in ecological networks holds great potential, especially for allowing rapid construction of large datasets of interactions, it is unclear how the choice of clustering threshold can influence the conclusions obtained. To test the impact of taxonomic precision on network architecture, we obtained and analyzed 16 datasets of ecological interactions, inferred from metabarcoding and observations. Our comparisons of networks constructed under a range of sequence thresholds for assigning taxa demonstrate that even small changes in node resolution can cause wide variation in almost all key metric values. Moreover, relative values of commonly used metrics such as robustness were seen to fluctuate continuously with node resolution, thereby potentially causing error in conclusions drawn when comparing multiple networks. In observational networks, we found that changing node resolution could, in some cases, lead to substantial changes to measurements of network topology. Overall, our findings highlight the importance of classifying nodes to the greatest precision possible, and demonstrate the need for caution when comparing networks that differ with respect to node resolution, even where taxonomic groups and interaction types are similar. In such cases, we recommend that comparisons of networks should focus on relative differences rather than absolute values between the networks studied.

中文翻译：

---

评估分类单元分辨率对网络结构的影响

构建生态网络已成为了解不同分类群如何相互作用的不可或缺的方法。然而，用于在网络研究中生成数据的方法因研究而异，这可能会限制我们有意义地比较结果的能力。特别是，节点分类方法的精度各不相同，可能会改变所研究网络的架构。例如，在观察研究中，分类群不是被归类为林奈物种，而是被定期分配给形态物种，或者在分子研究中被分配给分子操作分类单位 (MOTU)，后者是根据序列相似性的任意阈值定义的。尽管在生态网络中使用 MOTU 具有巨大的潜力，特别是在允许快速构建大型交互数据集方面，目前尚不清楚聚类阈值的选择如何影响所获得的结论。为了测试分类精度对网络架构的影响，我们获得并分析了 16 个生态相互作用数据集，从元条形码和观察中推断出来。我们对在分配分类群的一系列序列阈值下构建的网络的比较表明，即使节点分辨率的微小变化也会导致几乎所有关键度量值的广泛变化。此外，鲁棒性等常用指标的相对值会随着节点分辨率而不断波动，从而在比较多个网络时可能导致得出的结论出现错误。在观测网络中，我们发现在某些情况下，改变节点分辨率可能会导致网络拓扑测量的重大变化。全面的，我们的研究结果强调了尽可能精确地对节点进行分类的重要性，并证明在比较节点分辨率不同的网络时需要谨慎，即使分类组和交互类型相似。在这种情况下，我们建议网络的比较应该关注所研究的网络之间的相对差异而不是绝对值。