Fig. 5: MDSINE2 infers modular representations of complex microbiomial dynamical systems.
From: Learning ecosystem-scale dynamics from microbiome data with MDSINE2
Our method automatically learns modules of taxa (ASVs) on the basis of similarity of their dynamic interactions and responses to perturbations. Results are split into a, Gram-positive, and b, Gram-negative ASVs, for display purposes. a(i),b(i), ASV relative abundance. a(ii),b(ii), Phylogenetic tree of ASVs. a(iii),b(iii), Module memberships. Intensity of colour in the grid indicates abundance post colonization and before perturbations (average over days 14–21). c, Inferred module interaction network displaying only edges with BF > 100 (‘decisive’ evidence). Size of nodes is proportional to the number of ASVs in the module, and colour of nodes corresponds to module keystoneness. Num, number. d, Keystoneness analysis measures the relative importance of a module in maintaining the steady-state abundance of the community. A positive keystoneness for a module means that its removal from the community reduces the steady-state abundance of its community members, and a negative keystoneness means that the module’s removal results in an increase in the abundance of the other community members. A module’s impact on the other community members was computed both at the module level and across all modules (total keystoneness).
