Fig. 8

Network properties recapitulate physiology. a For each of the six clusters, which were defined from clustering scores in the multilevel sPLS model-fitted subspace, we show the treatment group identity (STAT/Control and NC/HFD, colored horizontal bars) and average physiological responses (vertical bar plots). Since each response is scaled and centered, the axes represent the mean response over the whole population at each time point. The state-change diagrams represent real-time transitions for the community in an individual mouse moving into a new cluster. For greater clarity, we removed transitions representing fewer than six mouse cluster changes. Clusters 1 and 3 are predominantly obtained from communities in STAT mice early-in-life, and Cluster 2 represents the early-in-life communities in control mice. The switch from NC to HFD corresponds to transitions from Cluster 3 to 5 and from Cluster 2 to 4. Transitions to Cluster 6 primarily include samples from week 30 STAT mice and week 18 and 30 Control mice. The circular arrows shown indicate those communities in mice that do not change clusters b We inferred networks using SPIEC-EASI [34] over the set of samples defined by each cluster. To compare graphs, we include a two-dimensional embedding of graphlet correlation distances (using isometric MDS, with the network positions shown as colored hexagons). These show that based on summarized local network topologies, closeness networks reflect cluster identity. The networks are shown in force directed layouts (overlaid on the ISOMDS, near their respective position in the embedding) and nodes are colored at the Phylum level, except for the two nodes with the highest betweenness (shown in gray, see also Additional file 6: Figure S6). c We used natural connectivity to assess the robustness of microbial ecological interaction networks to sequential node removals. The order of node removals was either random or ordered by degree or betweenness centrality. Natural connectivity is shown as a function of the relative size of the network