manta: a Clustering Algorithm for Weighted Ecological Networks

Abstract

Microbial network inference and analysis have become successful approaches to extract biological hypotheses from microbial sequencing data. Network clustering is a crucial step in this analysis. Here, we present a novel heuristic network clustering algorithm, manta, which clusters nodes in weighted networks. In contrast to existing algorithms, manta exploits negative edges while differentiating between weak and strong cluster assignments. For this reason, manta can tackle gradients and is able to avoid clustering problematic nodes. In addition, manta assesses the robustness of cluster assignment, which makes it more robust to noisy data than most existing tools. On noise-free synthetic data, manta equals or outperforms existing algorithms, while it identifies biologically relevant subcompositions in real-world data sets. On a cheese rind data set, manta identifies groups of taxa that correspond to intermediate moisture content in the rinds, while on an ocean data set, the algorithm identifies a cluster of organisms that were reduced in abundance during a transition period but did not correlate strongly to biochemical parameters that changed during the transition period. These case studies demonstrate the power of manta as a tool that identifies biologically informative groups within microbial networks. IMPORTANCE manta comes with unique strengths, such as the abilities to identify nodes that represent an intermediate between clusters, to exploit negative edges, and to assess the robustness of cluster membership. manta does not require parameter tuning, is straightforward to install and run, and can be easily combined with existing microbial network inference tools.

Open challenges for microbial network construction and analysis

“…Microbial network inference algorithms usually return "hairballs" of densely interconnected taxa that require further analysis to yield testable hypotheses. But despite the wealth of inference tools, only a few analysis tools dedicated to microbial networks have been developed to date [37][38][39]. Two types of network analysis are particularly informative: data integration and clustering.…”

“…Such taxon groups often covary in response to environmental factors such as pH and temperature, and therefore de novo clustering is a means to uncover niche structure. For instance, the members of different clusters in a cheese microbial network respond differently to moisture [37]. In the second case, taxa are assigned to clusters according to prior knowledge, for instance, their plankton function type [9] or their phylum.…”

Null-model-based network comparison reveals core associations

“…Therefore, taxa in the CAN were compared to taxa reported as indicators of high microbial abundance (HMA) or low microbial abundance (LMA) [32]. CAN network clusters were identified with manta v1.0.0 [33], as this algorithm has been designed to handle negative edges in the CAN. To run the clustering algorithm, default settings were used, except the number of iterations and permutations, which was set to 200.…”

Regime transition Shapes the Composition, Assembly Processes, and Co-occurrence Pattern of Bacterioplankton Community in a Large Eutrophic Freshwater Lake

“…Only correlations between environmental variables and species that were signi cant (P < 0.05, Benjamini-Hochberg adjust) and strong (r ≥ 0.75 or r ≤ -0.75) were considered. A novel microbial association network clustering algorithm (Manta) (version 1.0.1), which determines the optimal cluster number automatically, was used to separate the networks into clusters with default settings [22] after calculating the global network based on all 54 samples across the regime shift from MDR to PDR. Indexes, including modularity, clustering coe cient, average path length, network diameter, average degree and graph density were calculated using the 'igraph' packages in R to describe the attributes of a network [45].…”

“…This procedure demonstrated that both stochastic and deterministic processes appear to govern the assembly processes of bacterial communities [20,21]. However, a more complete understanding of community composition in the context of eutrophic lakes requires a thorough description of the abiotic drivers of co-occurrence patterns [22]. An inter-regime bacterioplankton co-occurrence network can demonstrate how clusters of organisms are driven by abiotic factors along regime shifts in freshwater lakes [23,24].…”