Identifying robust communities and multi-community nodes by combining top-down and bottom-up approaches to clustering

Abstract

Biological functions are carried out by groups of interacting molecules, cells or tissues, known as communities. Membership in these communities may overlap when biological components are involved in multiple functions. However, traditional clustering methods detect non-overlapping communities. These detected communities may also be unstable and difficult to replicate, because traditional methods are sensitive to noise and parameter settings. These aspects of traditional clustering methods limit our ability to detect biological communities, and therefore our ability to understand biological functions. To address these limitations and detect robust overlapping biological communities, we propose an unorthodox clustering method called SpeakEasy which identifies communities using top-down and bottom-up approaches simultaneously. Specifically, nodes join communities based on their local connections, as well as global information about the network structure. This method can quantify the stability of each community, automatically identify the number of communities, and quickly cluster networks with hundreds of thousands of nodes. SpeakEasy shows top performance on synthetic clustering benchmarks and accurately identifies meaningful biological communities in a range of datasets, including: gene microarrays, protein interactions, sorted cell populations, electrophysiology and fMRI brain imaging.

Deconvoluting simulated metagenomes: the performance of hard- and soft- clustering algorithms applied to metagenomic chromosome conformation capture (3C)

“…clique percolation, link partitioning, local expansion, fuzzy detection and label propagation). Label propagation methods have shown promise with respect to highly overlapped communities (Xie et al, 2011;Chen et al, 2010;Gaiteri et al, 2015), which we might reasonably expect to confront when resolving microbial strains. Empirically determined probability distributions, such as those governing the production of intra-chromosomal (cis) read-pairs as a function of genomic separation, might naturally lend themselves to methods from within the fuzzy-detection class.…”

Gene expression and DNA methylation are extensively coordinated with MRI-based brain microstructural characteristics

“…The gene coexpression methodology has been extended to DNA methylation, identifying loci and nearby genes whose methylation level fluctuate in sync, across many subjects (Numata et al 2012). To robustly identify coexpressed or comethylated gene sets, we use a consensus clustering method (Gaiteri et al 2015) that operates on the gene-gene Pearson correlation matrix (or CPG-CPG correlation matrix) to find gene sets whose expression or methylation levels covary across subjects. Average levels of these 47 gene and 58 methylation sets were then related to neuroimaging in the same cohort as described below.…”

Fuzzy overlapping community quality metrics

“…We also consider eight local community quality metrics: the number of Intra-edges, Intra-density, Contraction, the number of Boundary-edges, Expansion, Conductance [8,7], the Fitness function [24], and the Average Modularity Degree [25]. These metrics describe how the connectivity structure of a given set of nodes resembles a community.…”

“…In this paper, we consider several overlapping extensions of modularity and test their quality on real and synthetic networks. We also extend localized modularity [28], modularity density [8,7], and eight local community quality metrics for overlapping communities following the same principles used by the overlapping extensions of modularity.…”

“…We conducted experiments on a large number of real-world networks and synthetic networks using overlapping extensions of modularity, overlapping modularity density, and eight local metrics (the number of Intra-edges, Intradensity, Contraction, the number of Boundary-edges, Expansion, Conductance [8,7], the Fitness function [24], and the Average Modularity Degree [25]). The results show that selecting the product of the belonging coefficients of two nodes as a belonging function for overlapping extensions yields better results on these networks than using other belonging functions.…”