[Paper Review] PGLasso: Microbial Community Detection through Phylogenetic Graphical Lasso

Chieh Lo, Radu Mărculescu|arXiv (Cornell University)|Jul 20, 2018

Genomics and Phylogenetic Studies9 references4 citations

TL;DR

PGLasso proposes a phylogenetic graphical lasso method to infer microbial co-occurrence networks from metagenomic data by integrating phylogenetic information to address high-dimensionality and compositional data challenges. It outperforms GLasso, SparCC, and random null models with over 100% improvement in Adjusted Rand Index (ARI) on human microbiome datasets, demonstrating superior accuracy in detecting biologically meaningful microbial communities.

ABSTRACT

Due to the recent advances in high-throughput sequencing technologies, it becomes possible to directly analyze microbial communities in the human body and in the environment. Knowledge of how microbes interact with each other and form functional communities can provide a solid foundation to understand microbiome related diseases; this can serve as a key step towards precision medicine. In order to understand how microbes form communities, we propose a two step approach: First, we infer the microbial co-occurrence network by integrating a graph inference algorithm with phylogenetic information obtained directly from metagenomic data. Next, we utilize a network-based community detection algorithm to cluster microbes into functional groups where microbes in each group are highly correlated. We also curate a "gold standard" network based on the microbe-metabolic relationships which are extracted directly from the metagenomic data. Utilizing community detection on the resulting microbial metabolic pathway bipartite graph, the community membership for each microbe can be viewed as the true label when evaluating against other existing methods. Overall, our proposed framework Phylogenetic Graphical Lasso (PGLasso) outperforms existing methods with gains larger than 100% in terms of Adjusted Rand Index (ARI) which is commonly used to quantify the goodness of clusterings.

Motivation & Objective

• To address the limitations of traditional clustering methods in high-dimensional, compositional metagenomic data with unknown cluster counts.
• To improve microbial co-occurrence network inference by integrating phylogenetic information to reduce false positives and enhance biological relevance.
• To develop a robust framework for detecting functional microbial communities linked to microbiome-related diseases.
• To establish a gold standard for evaluation using microbe-metabolic pathway bipartite networks derived from metagenomic data.
• To evaluate the performance of network-based community detection against a curated, biologically grounded reference clustering.

Proposed method

• Pre-process metagenomic data to extract microbial abundance and microbe-metabolic pathway relationships using the Humann2 pipeline.
• Construct phylogenetic trees using genome sequences from PATRIC for each microbe, leveraging PhyloPhlAn for accurate evolutionary relationships.
• Apply a modified graphical lasso algorithm (PGLasso) that incorporates phylogenetic distance as a regularization prior to infer microbial co-occurrence networks.
• Use the Louvain algorithm for modularity maximization to detect communities in the inferred co-occurrence network.
• Construct a gold standard bipartite network of microbes and metabolic pathways to serve as ground truth for clustering evaluation.
• Evaluate clustering performance using Adjusted Rand Index (ARI), correcting for chance agreement between inferred and true clusterings.

Experimental results

Research questions

• RQ1 Can integrating phylogenetic information into graphical lasso improve the accuracy of microbial co-occurrence network inference in high-dimensional, compositional metagenomic data?
• RQ2 How does PGLasso compare to established methods like GLasso and SparCC in detecting biologically meaningful microbial communities?
• RQ3 To what extent does the use of a curated microbe-metabolic pathway bipartite network as a gold standard improve the evaluation of community detection algorithms?
• RQ4 Does the incorporation of phylogenetic priors lead to more stable and biologically plausible clustering results compared to random or unregularized methods?
• RQ5 Can the PGLasso framework generalize to different human body sites and detect consistent microbial community structures across diverse niches?

Key findings

• PGLasso achieved a 100%+ improvement in Adjusted Rand Index (ARI) compared to baseline methods across all tested body sites, with the highest ARI of 0.303 on the anterior nares (AntNar) dataset.
• The method significantly outperformed GLasso, SparCC, and random null models, with ARI values consistently above 0.1 on all datasets, indicating strong clustering accuracy.
• The incorporation of phylogenetic priors reduced the solution space, leading to more accurate network inference and better alignment with the gold standard metabolic pathway network.
• SparCC and GLasso showed poor performance, with ARI values near or below 0.05, indicating clustering results were nearly indistinguishable from random.
• The Louvain community detection algorithm successfully identified biologically coherent microbial groups in the PGLasso-inferred networks, suggesting functional coherence.
• The gold standard metabolic pathway bipartite network provided a reliable benchmark, confirming that PGLasso’s inferred communities are more biologically plausible than those from other methods.