The wealth of sequencing data now available for microbial communities has led to a blossoming of interest in their dynamics and functional roles in ecosystem processes. However, addressing these topics is a major analytical challenge due to the size of the communities, uneven abundance distributions, significant stochasticity and the rapidity of dynamics. Overcoming these challenges is particularly important as methodologies developed for microbial communities are likely to form the spearhead of techniques which can be used to address similar topics across a wide range of other community types, as the advances in technology (e.g. eDNA) lead to similar high resolution data being collected for a broader range of systems.
Methods based on network analysis have demonstrated substantial promise for developing our understanding of microbial community dynamics. These methods have helped to both quantify and visualise various aspects of community structure. More recent developments have started to focus on temporal changes in communities in order to identify key driving species, or clusters of core species interacting with each other. However, the impact of the sampling design on the robustness of these network metrics, and therefore the conclusion drawn from them, has received little attention.
Results/Conclusions
We start by illustrating a novel network analytical approach using data from a study on microbial communities associated with healthy corals and corals infected with a new disease (grey-patch disease) found in Micronesia. Importantly, the network analysis identified how microbial clusters rather than single pathogenic agents are likely to be driving the degradation of the coral defences, leading to disease progression. Using computer simulations we then examine how robust this methodology, and several recently developed metrics focussing on network dynamics, are to variation in sampling design. We illustrate that the temporal resolution of sampling can have profound impacts on the conclusions drawn. In particular, the sampling frequency relative to the speed of community dynamics, the extent of stochastic variation, and the number of samples taken can strongly reduce the reliability of identifying key species or clusters driving community dynamics. We conclude by showing how using samples taken from multiple locations at the same time and using network analyses to infer community dynamics can lead to erroneous conclusions.