A spectrum of gut microbiota signatures predicts long-term health status

Figure 1. Overview of latent factorization into gut microbiota signatures. Schematic illustration of a latent factorization of fecal microbiota profiles into three sub-communities. The observed microbial abundance data table (left) is represented as a mixture of three latent signatures (ES). Each individual has a unique mixed membership of the signatures with varying weights (middle), and each signature corresponds to a specific community profile (right).

Figure 2. Both standard and high-granularity signatures were associated with mortality and common morbidities. (A) Pairwise cross-sectional correlation of the signatures (k = 5) with significant health variables. (B) k = 5 granularity level. Four multivariate Cox regression models for all-cause mortality, incident diabetes, liver disease, and sepsis responses. An upward-pointing triangle represents HR > 1, and a downward-pointing triangle represents HR < 1. (C) k = 5 granularity level. Kaplan-Meier (KM) analysis for categorized all-cause mortality response, where ES-Bact is set as the reference level. Only significant (q < 0.05) principal ESs are plotted. (D) k = 20 granularity level. Four multivariate Cox regression models for all-cause mortality, diabetes, liver disease, and sepsis responses. (E) k = 20 granularity level. KM analysis for categorized all-cause mortality response, where ES-Bact. H uniformis is set as the reference level. Only significant principal ESs are plotted.

Figure 3. Granularity-robust similarity captures the features of the mortality-associated paths across the alignment. (A) The relative composition of each of the 60 unique indicator species combined both in the standard (k = 5) and high-granularity (k = 20) models. Circle size refers to the magnitude of each species’ contribution to the ES. (B) Structure of component alignment across 20 models (k ∈ {1,2,…,20}), where the nodes represent each ES in each model, and the width of the edges represents the similarity (inner product) of the signatures between k and k+1. Illustrates how the signatures evolve with k. Statistically significant results from multivariate Cox regression models for all-cause mortality response. (C) Relative similarity between the standard (k = 5) and high-granularity (k = 20) models, representing the divergence of the signatures. Emphasizes that eubiotic ES-Alis (k = 5) is a parent of many signatures including ES-Alis. A putredinis and ES-RUG115, as well as dysbiotic ES-Esch, is a parent signature of ES-Esch. ryusiae and ES-Flav. plautii, and ES-Prev primarily of ES-Prev. copri.