A Comparison of Community Detection Algorithms on Artificial Networks

From attributes to communities: a novel approach in social network generation

Generating networks with attributes would be useful in computer game development by enabling dynamic social interactions, adaptive storylines, realistic economic systems, ecosystem modelling, urban development, strategic planning, and adaptive learning systems. To this end, we propose the Attribute-based Realistic Community and Associate NEtwork (ARCANE) algorithm to generate node-attributed networks with functional communities. We have designed a numerical node attribute-edge relationship computation system to handle the edge generation phase of our network generator, which is a different method from our predecessors. We combine this system with the proximity between nodes to create more life-like communities. Our method is compared against other node-attributed social network generators in the area with using both different evaluation metrics and a real-world dataset. The model properties evaluation identified ARCANE as the leading generator, with another generator ranking in a tie for first place. As a more favorable outcome for our approach, the community detection evaluation indicated that ARCANE exhibited superior performance compared to other competing generators within this domain. This thorough evaluation of the resulting graphs show that the proposed method can be an alternate approach to social network generators with node attributes and communities.

On the role of diffusion dynamics on community-aware centrality measures

Theoretical and empirical studies on diffusion models have revealed their versatile applicability across different fields, spanning from sociology and finance to biology and ecology. The presence of a community structure within real-world networks has a substantial impact on how diffusion processes unfold. Key nodes located both within and between these communities play a crucial role in initiating diffusion, and community-aware centrality measures effectively identify these nodes. While numerous diffusion models have been proposed in literature, very few studies investigate the relationship between the diffusive ability of key nodes selected by community-aware centrality measures, the distinct dynamical conditions of various models, and the diverse network topologies. By conducting a comparative evaluation across four diffusion models, utilizing both synthetic and real-world networks, along with employing two different community detection techniques, our study aims to gain deeper insights into the effectiveness and applicability of the community-aware centrality measures. Results suggest that the diffusive power of the selected nodes is affected by three main factors: the strength of the network's community structure, the internal dynamics of each diffusion model, and the budget availability. Specifically, within the category of simple contagion models, such as SI, SIR, and IC, we observe similar diffusion patterns when the network's community structure strength and budget remain constant. In contrast, the LT model, which falls under the category of complex contagion dynamics, exhibits divergent behavior under the same conditions.

Identification of Real-Life Mixtures Using Human Biomonitoring Data: A Proof of Concept Study

Human health risk assessment of chemical mixtures is complex due to the almost infinite number of possible combinations of chemicals to which people are exposed to on a daily basis. Human biomonitoring (HBM) approaches can provide inter alia information on the chemicals that are in our body at one point in time. Network analysis applied to such data may provide insight into real-life mixtures by visualizing chemical exposure patterns. The identification of groups of more densely correlated biomarkers, so-called "communities", within these networks highlights which combination of substances should be considered in terms of real-life mixtures to which a population is exposed. We applied network analyses to HBM datasets from Belgium, Czech Republic, Germany, and Spain, with the aim to explore its added value for exposure and risk assessment. The datasets varied in study population, study design, and chemicals analysed. Sensitivity analysis was performed to address the influence of different approaches to standardise for creatinine content of urine. Our approach demonstrates that network analysis applied to HBM data of highly varying origin provides useful information with regards to the existence of groups of biomarkers that are densely correlated. This information is relevant for regulatory risk assessment, as well as for the design of relevant mixture exposure experiments.

Detecting overlapping communities in complex networks using non-cooperative games

Detecting communities in complex networks is of paramount importance, and its wide range of real-life applications in various areas has caused a lot of attention to be paid to it, and many efforts have been made to have efficient and accurate algorithms for this purpose. In this paper, we proposed a non-cooperative game theoretic-based algorithm that is able to detect overlapping communities. In this algorithm, nodes are regarded as players, and communities are assumed to be groups of players with similar strategies. Our two-phase algorithm detects communities and the overlapping nodes in separate phases that, while increasing the accuracy, especially in detecting overlapping nodes, brings about higher algorithm speed. Moreover, there is no need for setting parameters regarding the size or number of communities, and the absence of any stochastic process caused this algorithm to be stable. By appropriately adjusting stop criteria, our algorithm can be categorized among those with linear time complexity, making it highly scalable for large networks. Experiments on synthetic and real-world networks demonstrate our algorithm's good performance compared to similar algorithms in terms of detected overlapping nodes, detected communities size distribution, modularity, and normalized mutual information.

Understanding Pain and Trauma Symptoms in Veterans From Resting-State Connectivity: Unsupervised Modeling

Trauma and posttraumatic stress are highly comorbid with chronic pain and are often antecedents to developing chronic pain conditions. Pain and trauma are associated with greater utilization of medical services, greater use of psychiatric medication, and increased total cost of treatment. Despite the high overlap in the clinic, the neural mechanisms of pain and trauma are often studied separately. In this study, resting-state functional magnetic resonance imaging (rs-fMRI) scans were completed among a diagnostically heterogeneous sample of veterans with a range of back pain and trauma symptoms. Using Group Iterative Multiple Model Estimation (GIMME), an effective functional connectivity analysis, we explored an unsupervised model deriving subgroups based on path similarity in a priori defined regions of interest (ROIs) from brain regions implicated in the experience of pain and trauma. Three subgroups were identified by patterns in functional connection and differed significantly on several psychological measures despite similar demographic and diagnostic characteristics. The first subgroup was highly connected overall, was characterized by functional connectivity from the nucleus accumbens (NAc), the anterior cingulate cortex (ACC), and the posterior cingulate cortex (PCC) to the insula and scored low on pain and trauma symptoms. The second subgroup did not significantly differ from the first subgroup on pain and trauma measures but was characterized by functional connectivity from the ACC and NAc to the thalamus and from ACC to PCC. The third subgroup was characterized by functional connectivity from the thalamus and PCC to NAc and scored high on pain and trauma symptoms. Our results suggest that, despite demographic and diagnostic similarities, there may be neurobiologically dissociable biotypes with different mechanisms for managing pain and trauma. These findings may have implications for the determination of appropriate biotype-specific interventions that target these neurological systems.

Neural heterogeneity underlying late adolescent motivational processing is linked to individual differences in behavioral sensation seeking

Adolescent risk-taking, including sensation seeking (SS), is often attributed to developmental changes in connectivity among brain regions implicated in cognitive control and reward processing. Despite considerable scientific and popular interest in this neurodevelopmental framework, there are few empirical investigations of adolescent functional connectivity, let alone examinations of its links to SS behavior. The studies that have been done focus on mean-based approaches and leave unanswered questions about individual differences in neurodevelopment and behavior. The goal of this paper is to take a person-specific approach to the study of adolescent functional connectivity during a continuous motivational state, and to examine links between connectivity and self-reported SS behavior in 104 adolescents (MAge  = 19.3; SDAge  = 1.3). Using Group Iterative Multiple Model Estimation (GIMME), person-specific connectivity during two neuroimaging runs of a monetary incentive delay task was estimated among 12 a priori brain regions of interest representing reward, cognitive, and salience networks. Two data-driven subgroups were detected, a finding that was consistent between both neuroimaging runs, but associations with SS were only found in the first run, potentially reflecting neural habituation in the second run. Specifically, the subgroup that had unique connections between reward-related regions had greater SS and showed a distinctive relation between connectivity strength in the reward regions and SS. These findings provide novel evidence for heterogeneity in adolescent brain-behavior relations by showing that subsets of adolescents have unique associations between neural motivational processing and SS. Findings have broader implications for future work on reward processing, as they demonstrate that brain-behavior relations may attenuate across runs.

Synwalk: community detection via random walk modelling

Complex systems, abstractly represented as networks, are ubiquitous in everyday life. Analyzing and understanding these systems requires, among others, tools for community detection. As no single best community detection algorithm can exist, robustness across a wide variety of problem settings is desirable. In this work, we present Synwalk, a random walk-based community detection method. Synwalk builds upon a solid theoretical basis and detects communities by synthesizing the random walk induced by the given network from a class of candidate random walks. We thoroughly validate the effectiveness of our approach on synthetic and empirical networks, respectively, and compare Synwalk’s performance with the performance of Infomap and Walktrap (also random walk-based), Louvain (based on modularity maximization) and stochastic block model inference. Our results indicate that Synwalk performs robustly on networks with varying mixing parameters and degree distributions. We outperform Infomap on networks with high mixing parameter, and Infomap and Walktrap on networks with many small communities and low average degree. Our work has a potential to inspire further development of community detection via synthesis of random walks and we provide concrete ideas for future research.

The network of psychosocial health in middle-aged and older adults during the first COVID-19 lockdown

PURPOSE

Psychosocial health problems, such as social isolation, loneliness, depression and anxiety, have gained attention during the COVID-19 pandemic and are commonly co-occurring. We investigated the network of psychosocial health constructs during the COVID-19 pandemic.

METHODS

This study included 4553 participants (mean age: 68.6 ± 11.2 years, 56% women) from the prospective Rotterdam Study, who filled out a questionnaire between April and July 2020, the time of the first COVID-19 wave in the Netherlands. Psychosocial health constructs included were depressive symptoms (Center for Epidemiological Studies Depression scale), anxiety symptoms (Hospital Anxiety and Depression scale), loneliness (University of California, Los Angeles loneliness scale), social connectedness (five items) and pandemic-related worry (five items). We estimated mixed graphical models to assess the network of items of these constructs and whether age and sex affected the network structure.

RESULTS

Within the network of psychosocial constructs, a higher depressive symptoms score was particularly associated with items of loneliness and social connectedness, whereas overall anxiety was particularly associated with items of pandemic-related worry. Between people from different sex and age, the network structure significantly altered.

CONCLUSION

This study demonstrates that within the same network of psychosocial health constructs, depressive symptom score is particularly associated with loneliness and social connectedness, whereas anxiety symptom score is associated with pandemic-related worry during the first COVID-19 lockdown. Our results support that psychosocial constructs should be considered in conjunction with one another in prevention and treatment efforts in clinical care, and that these efforts need to be tailored to specific demographic groups.

Systematic Review of Functional MRI Applications for Psychiatric Disease Subtyping

Background: Psychiatric disorders have been historically classified using symptom information alone. Recently, there has been a dramatic increase in research interest not only in identifying the mechanisms underlying defined pathologies but also in redefining their etiology. This is particularly relevant for the field of personalized medicine, which searches for data-driven approaches to improve diagnosis, prognosis, and treatment selection for individual patients. Methods: This review aims to provide a high-level overview of the rapidly growing field of functional magnetic resonance imaging (fMRI) from the perspective of unsupervised machine learning applications for disease subtyping. Following the PRISMA guidelines for protocol reproducibility, we searched the PubMed database for articles describing functional MRI applications used to obtain, interpret, or validate psychiatric disease subtypes. We also employed the active learning framework ASReview to prioritize publications in a machine learning-guided way. Results: From the 20 studies that met the inclusion criteria, five used functional MRI data to interpret symptom-derived disease clusters, four used it to interpret clusters derived from biomarker data other than fMRI itself, and 11 applied clustering techniques involving fMRI directly. Major depression disorder and schizophrenia were the two most frequently studied pathologies (35% and 30% of the retrieved studies, respectively), followed by ADHD (15%), psychosis as a whole (10%), autism disorder (5%), and the consequences of early exposure to violence (5%). Conclusions: The increased interest in personalized medicine and data-driven disease subtyping also extends to psychiatric disorders. However, to date, this subfield is at an incipient exploratory stage, and all retrieved studies were mostly proofs of principle where further validation and increased sample sizes are craved for. Whereas results for all explored diseases are inconsistent, we believe this reflects the need for concerted, multisite data collection efforts with a strong focus on measuring the generalizability of results. Finally, whereas functional MRI is the best way of measuring brain function available to date, its low signal-to-noise ratio and elevated monetary cost make it a poor clinical alternative. Even with technology progressing and costs decreasing, this might incentivize the search for more accessible, clinically ready functional proxies in the future.

Unfolding the dynamical structure of Lisbon's public space: space syntax and micromobility data

Space Syntax and the theory of natural movement demonstrated that spatial morphology is a primary factor influencing movement. This paper investigates to what extent spatial morphology at different scales (node, community and global network) influences the use of public space by micromobility. An axial map and corresponding network for Lisbon's walkable and open public space, and data from e-scooters parking locations, is used as case study. Relevant metrics and their correlations (intelligibility, accessibility, permeability and local dimension) for the quantitative characterization of spatial morphology properties are described and computed for Lisbon's axial map. Communities are identified based on the network topological structure in order to investigate how these properties are affected at different scales in the case study. The resulting axial line clustering is compared via the variation of information metric with the clustering obtained from e-scooters' proximity. The results obtained enable to conclude that the space syntax properties are scale dependent in Lisbon's pedestrian network. On the other hand both the correlation between these properties, the number of scooters and the variation of information between clusters indicate that the spatial morphology is not the only factor influencing micromobility. Through the comparative analysis between the main properties of the public space network of Lisbon and data collected from e-scooters locations in a timeframe, centrality becomes a dynamic concept, relying not only on the static topological properties of the urban network, but also on other quantitative and qualitative factors, since the flows' operating on the network will operate several transformations on the spatial network properties through time, uncovering spatiotemporal dynamics.

Comparing Graph Clusterings: Set Partition Measures vs. Graph-Aware Measures

In this paper, we propose a family of graph partition similarity measures that take the topology of the graph into account. These graph-aware measures are alternatives to using set partition similarity measures that are not specifically designed for graphs. The two types of measures, graph-aware and set partition measures, are shown to have opposite behaviors with respect to resolution issues and provide complementary information necessary to compare graph partitions.

Network Analysis to Identify Communities Among Multiple Exposure Biomarkers Measured at Birth in Three Flemish General Population Samples

Introduction: Humans are exposed to multiple environmental chemicals via different sources resulting in complex real-life exposure patterns. Insight into these patterns is important for applications such as linkage to health effects and (mixture) risk assessment. By providing internal exposure levels of (metabolites of) chemicals, biomonitoring studies can provide snapshots of exposure patterns and factors that drive them. Presentation of biomonitoring data in networks facilitates the detection of such exposure patterns and allows for the systematic comparison of observed exposure patterns between datasets and strata within datasets. Methods: We demonstrate the use of network techniques in human biomonitoring data from cord blood samples collected in three campaigns of the Flemish Environment and Health Studies (FLEHS) (sampling years resp. 2002-2004, 2008-2009, and 2013-2014). Measured biomarkers were multiple organochlorine compounds, PFAS and metals. Comparative network analysis (CNA) was conducted to systematically compare networks between sampling campaigns, smoking status during pregnancy, and maternal pre-pregnancy BMI. Results: Network techniques offered an intuitive approach to visualize complex correlation structures within human biomonitoring data. The identification of groups of highly connected biomarkers, "communities," within these networks highlighted which biomarkers should be considered collectively in the analysis and interpretation of epidemiological studies or in the design of toxicological mixture studies. Network analyses demonstrated in our example to which extent biomarker networks and its communities changed across the sampling campaigns, smoking status during pregnancy, and maternal pre-pregnancy BMI. Conclusion: Network analysis is a data-driven and intuitive screening method when dealing with multiple exposure biomarkers, which can easily be upscaled to high dimensional HBM datasets, and can inform mixture risk assessment approaches.

Network Analysis of the Multidimensional Symptom Experience of Oncology

Oncology patients undergoing cancer treatment experience an average of fifteen unrelieved symptoms that are highly variable in both their severity and distress. Recent advances in Network Analysis (NA) provide a novel approach to gain insights into the complex nature of co-occurring symptoms and symptom clusters and identify core symptoms. We present findings from the first study that used NA to examine the relationships among 38 common symptoms in a large sample of oncology patients undergoing chemotherapy. Using two different models of Pairwise Markov Random Fields (PMRF), we examined the nature and structure of interactions for three different dimensions of patients’ symptom experience (i.e., occurrence, severity, distress). Findings from this study provide the first direct evidence that the connections between and among symptoms differ depending on the symptom dimension used to create the network. Based on an evaluation of the centrality indices, nausea appears to be a structurally important node in all three networks. Our findings can be used to guide the development of symptom management interventions based on the identification of core symptoms and symptom clusters within a network.

Data-Driven Subgroups in Depression Derived from Directed Functional Connectivity Paths at Rest

Depressed patients show abnormalities in brain connectivity at rest, including hyperconnectivity within the default mode network (DMN). However, there is well-known heterogeneity in the clinical presentation of depression that is overlooked when averaging connectivity data. We used data-driven parsing of neural connectivity to reveal subgroups among 80 depressed patients completing resting state fMRI. Directed functional connectivity paths (eg, region A influences region B) within a depression-relevant network were characterized using Group Iterative Multiple Model Estimation, a method shown to accurately recover the direction and presence of connectivity paths in individual participants. Individuals were clustered using community detection on neural connectivity estimates. Subgroups were compared on network features and on clinical and biological/demographic characteristics that influence depression prognosis. Two subgroups emerged. Subgroup A, containing 71% of the patients, showed a typical pattern of connectivity across DMN nodes, as previously reported in depressed patients on average. Subgroup B exhibited an atypical connectivity profile lacking DMN connectivity, with increased dorsal anterior cingulate-driven connectivity paths. Subgroup B members had an over-representation of females (87% of Subgroup B vs 65% of Subgroup A; χ²=3.89, p=0.049), comorbid anxiety diagnoses (42.9% of Subgroup B vs 17.5% of Subgroup A; χ²=5.34, p=.02), and highly recurrent depression (63.2% of Subgroup B vs 31.8% of Subgroup A; χ²=5.38, p=.02). Neural connectivity-based categorization revealed an atypical pattern of connectivity in a depressed patient subset that would be overlooked in group comparisons of depressed and healthy participants, and tracks with clinically relevant phenotypes including anxious depression and episodic recurrence. Data-driven parsing suggests heterogeneous substrates of depression; ideally, future work building on these findings will inform personalized treatment.

Unsupervised Classification During Time-Series Model Building

Researchers who collect multivariate time-series data across individuals must decide whether to model the dynamic processes at the individual level or at the group level. A recent innovation, group iterative multiple model estimation (GIMME), offers one solution to this dichotomy by identifying group-level time-series models in a data-driven manner while also reliably recovering individual-level patterns of dynamic effects. GIMME is unique in that it does not assume homogeneity in processes across individuals in terms of the patterns or weights of temporal effects. However, it can be difficult to make inferences from the nuances in varied individual-level patterns. The present article introduces an algorithm that arrives at subgroups of individuals that have similar dynamic models. Importantly, the researcher does not need to decide the number of subgroups. The final models contain reliable group-, subgroup-, and individual-level patterns that enable generalizable inferences, subgroups of individuals with shared model features, and individual-level patterns and estimates. We show that integrating community detection into the GIMME algorithm improves upon current standards in two important ways: (1) providing reliable classification and (2) increasing the reliability in the recovery of individual-level effects. We demonstrate this method on functional MRI from a sample of former American football players.

A Comparative Analysis of Community Detection Algorithms on Artificial Networks

Many community detection algorithms have been developed to uncover the mesoscopic properties of complex networks. However how good an algorithm is, in terms of accuracy and computing time, remains still open. Testing algorithms on real-world network has certain restrictions which made their insights potentially biased: the networks are usually small, and the underlying communities are not defined objectively. In this study, we employ the Lancichinetti-Fortunato-Radicchi benchmark graph to test eight state-of-the-art algorithms. We quantify the accuracy using complementary measures and algorithms' computing time. Based on simple network properties and the aforementioned results, we provide guidelines that help to choose the most adequate community detection algorithm for a given network. Moreover, these rules allow uncovering limitations in the use of specific algorithms given macroscopic network properties. Our contribution is threefold: firstly, we provide actual techniques to determine which is the most suited algorithm in most circumstances based on observable properties of the network under consideration. Secondly, we use the mixing parameter as an easily measurable indicator of finding the ranges of reliability of the different algorithms. Finally, we study the dependency with network size focusing on both the algorithm's predicting power and the effective computing time.

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

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.

A new method for constructing networks from binary data

Network analysis is entering fields where network structures are unknown, such as psychology and the educational sciences. A crucial step in the application of network models lies in the assessment of network structure. Current methods either have serious drawbacks or are only suitable for Gaussian data. In the present paper, we present a method for assessing network structures from binary data. Although models for binary data are infamous for their computational intractability, we present a computationally efficient model for estimating network structures. The approach, which is based on Ising models as used in physics, combines logistic regression with model selection based on a Goodness-of-Fit measure to identify relevant relationships between variables that define connections in a network. A validation study shows that this method succeeds in revealing the most relevant features of a network for realistic sample sizes. We apply our proposed method to estimate the network of depression and anxiety symptoms from symptom scores of 1108 subjects. Possible extensions of the model are discussed.