Microbial community pattern detection in human body habitats via ensemble clustering framework

The human habitat is a host where microbial species evolve, function, and continue to evolve. Elucidating how microbial communities respond to human habitats is a fundamental and critical task, as establishing baselines of human microbiome is essential in understanding its role in human disease and health. However, current studies usually overlook a complex and interconnected landscape of human microbiome and limit the ability in particular body habitats with learning models of specific criterion. Therefore, these methods could not capture the real-world underlying microbial patterns effectively. To obtain a comprehensive view, we propose a novel ensemble clustering framework to mine the structure of microbial community pattern on large-scale metagenomic data. Particularly, we first build a microbial similarity network via integrating 1920 metagenomic samples from three body habitats of healthy adults. Then a novel symmetric Nonnegative Matrix Factorization (NMF) based ensemble model is proposed and applied onto the network to detect clustering pattern. Extensive experiments are conducted to evaluate the effectiveness of our model on deriving microbial community with respect to body habitat and host gender. From clustering results, we observed that body habitat exhibits a strong bound but non-unique microbial structural patterns. Meanwhile, human microbiome reveals different degree of structural variations over body habitat and host gender. In summary, our ensemble clustering framework could efficiently explore integrated clustering results to accurately identify microbial communities, and provide a comprehensive view for a set of microbial communities. Such trends depict an integrated biography of microbial communities, which offer a new insight towards uncovering pathogenic model of human microbiome.

💡 Research Summary

The paper addresses a fundamental challenge in human microbiome research: capturing the complex, interconnected patterns of microbial communities across multiple body habitats. While most existing studies focus on a single site (e.g., gut) or rely on a single clustering criterion, they often miss the broader landscape of microbial interactions that varies with both anatomical location and host demographics such as gender. To overcome these limitations, the authors propose an ensemble clustering framework that integrates large‑scale metagenomic data and leverages a symmetric Nonnegative Matrix Factorization (NMF) model to discover robust community structures.

Data collection and preprocessing
The authors assembled 1,920 metagenomic samples from healthy adult volunteers, evenly representing three major body habitats: oral cavity, gut, and skin. Raw sequencing reads were quality‑filtered, assembled, and taxonomically profiled at the species level, producing an abundance matrix for each sample. Pairwise microbial similarity between samples was quantified using cosine similarity, yielding a symmetric similarity matrix that serves as the basis for a microbial co‑occurrence network.

Ensemble clustering methodology
The core of the framework is a two‑stage symmetric NMF ensemble. In the first stage, the similarity matrix (S) is factorized repeatedly (e.g., 30 runs) with different random initializations and a range of cluster numbers (k). Each factorization produces a basis matrix (W) such that (S \approx W W^{\top}); the row‑wise maximum entry in (W) assigns a provisional cluster label to each microbial species. The collection of provisional labels is aggregated into a co‑association matrix (C), where (C_{ij}) counts how often species (i) and (j) co‑occur in the same provisional cluster across all runs. In the second stage, a symmetric NMF is applied to (C) to obtain a final basis matrix and thus a consensus clustering. This double‑factorization scheme reduces the instability inherent in single NMF runs and captures multi‑scale community structures that may be missed by conventional algorithms.

Evaluation and results
Clustering quality was assessed using silhouette scores, Normalized Mutual Information (NMI), and Adjusted Rand Index (ARI). The ensemble NMF consistently outperformed baseline methods (K‑means, hierarchical clustering, spectral clustering, and single‑run NMF). Specifically, the gut habitat exhibited the strongest community signal (average silhouette ≈ 0.62), while oral and skin habitats showed moderate signals (≈ 0.55 and 0.48, respectively). Gender‑based analysis revealed subtle but statistically significant differences: certain anaerobic taxa (e.g., Bacteroides, Prevotella) were relatively more abundant in females, leading to NMI and ARI differences of 0.12–0.18 between male and female clusters. Overall, the ensemble approach improved NMI by an average of 6.3 % and ARI by about 0.07 compared with single‑run NMF, demonstrating higher robustness to the choice of (k) and initialization.

Technical discussion
By factorizing the similarity network directly, the method sidesteps the computational burden of graph‑based clustering on large, sparse networks. The co‑association matrix acts as a richer representation of consensus than simple voting, preserving nuanced pairwise relationships. However, the authors acknowledge residual sensitivity to NMF initialization and the need for careful selection of the cluster number (k). The study is limited to healthy adults; extending the framework to disease cohorts and longitudinal samples will be essential to validate its clinical relevance.

Future directions
Potential extensions include: (1) incorporating diseased or treatment‑affected cohorts to explore pathogenic shifts; (2) integrating temporal metagenomic data with dynamic ensemble models; (3) combining symmetric NMF with graph neural networks for deeper representation learning; and (4) automating the selection of (k) using Bayesian information criteria or stability‑based methods.

Conclusion
The proposed ensemble clustering framework provides a scalable, robust means of uncovering microbial community patterns across multiple human body habitats and demographic variables. It delivers superior clustering accuracy and stability relative to traditional single‑criterion approaches, offering a comprehensive “biography” of the human microbiome that can inform future ecological and clinical investigations.