Lower Bounds on Paraclique Density

A Comparative Study of Gene Co-Expression Thresholding Algorithms

The thresholding problem is studied in the context of graph theoretical analysis of gene co-expression data. A number of thresholding methodologies are described, implemented, and tested over a large collection of graphs derived from real high-throughput biological data. Comparative results are presented and discussed.

Seminar: Scalable Preprocessing Tools for Exposomic Data Analysis

BACKGROUND

The exposome serves as a popular framework in which to study exposures from chemical and nonchemical stressors across the life course and the differing roles that these exposures can play in human health. As a result, data relevant to the exposome have been used as a resource in the quest to untangle complicated health trajectories and help connect the dots from exposures to adverse outcome pathways.

OBJECTIVES

The primary aim of this methods seminar is to clarify and review preprocessing techniques critical for accurate and effective external exposomic data analysis. Scalability is emphasized through an application of highly innovative combinatorial techniques coupled with more traditional statistical strategies. The Public Health Exposome is used as an archetypical model. The novelty and innovation of this seminar's focus stem from its methodical, comprehensive treatment of preprocessing and its demonstration of the positive effects preprocessing can have on downstream analytics.

DISCUSSION

State-of-the-art technologies are described for data harmonization and to mitigate noise, which can stymie downstream interpretation, and to select key exposomic features, without which analytics may lose focus. A main task is the reduction of multicollinearity, a particularly formidable problem that frequently arises from repeated measurements of similar events taken at various times and from multiple sources. Empirical results highlight the effectiveness of a carefully planned preprocessing workflow as demonstrated in the context of more highly concentrated variable lists, improved correlational distributions, and enhanced downstream analytics for latent relationship discovery. The nascent field of exposome science can be characterized by the need to analyze and interpret a complex confluence of highly inhomogeneous spatial and temporal data, which may present formidable challenges to even the most powerful analytical tools. A systematic approach to preprocessing can therefore provide an essential first step in the application of modern computer and data science methods. https://doi.org/10.1289/EHP12901.

Building effective intervention models utilizing big data to prevent the obesity epidemic

INTRODUCTION

The exposome consists of factors an individual is exposed to across the life course. The exposome is dynamic, meaning the factors are constantly changing, affecting each other and individuals in different ways. Our exposome dataset includes social determinants of health as well as policy, climate, environment, and economic factors that could impact obesity development. The objective was to translate spatial exposure to these factors with the presence of obesity into actionable population-based constructs that could be further explored.

METHODS

Our dataset was constructed from a combination of public-use datasets and the Center of Disease Control's Compressed Mortality File. Spatial Statistics using Queens First Order Analysis was performed to identify hot- and cold-spots of obesity prevalence; followed by Graph Analysis, Relational Analysis, and Exploratory Factor Analysis to model the multifactorial spatial connections.

RESULTS

Areas of high and low presence of obesity had different factors associated with obesity. Factors associated with obesity in areas of high obesity propensity were: poverty / unemployment; workload, comorbid conditions (diabetes, CVD) and physical activity. Conversely, factors associated in areas where obesity was rare were: smoking, lower education, poorer mental health, lower elevations, and heat.

DISCUSSION

The spatial methods described within the paper are scalable to large numbers of variables without issues of multiple comparisons lowering resolution. These types of spatial structural methods provide insights into novel variable associations or factor interactions that can then be studied further at the population or policy levels.

The public health exposome and pregnancy-related mortality in the United States: a high-dimensional computational analysis

BACKGROUND

Racial inequities in maternal mortality in the U.S. continue to be stark.

METHODS

The 2015-2018, 4-year total population, county-level, pregnancy-related mortality ratio (PRM; deaths per 100,000 live births; National Center for Health Statistics (NCHS), restricted use mortality file) was linked with the Public Health Exposome (PHE). Using data reduction techniques, 1591 variables were extracted from over 62,000 variables for use in this analysis, providing information on the relationships between PRM and the social, health and health care, natural, and built environments. Graph theoretical algorithms and Bayesian analysis were applied to PHE/PRM linked data to identify latent networks.

RESULTS

PHE variables most strongly correlated with total population PRM were years of potential life lost and overall life expectancy. Population-level indicators of PRM were overall poverty, smoking, lack of exercise, heat, and lack of adequate access to food.

CONCLUSIONS

In this high-dimensional analysis, overall life expectancy, poverty indicators, and health behaviors were found to be the strongest predictors of pregnancy-related mortality. This provides strong evidence that maternal death is part of a broader constellation of both similar and unique health behaviors, social determinants and environmental exposures as other causes of death.

Molecular Subtyping and Outlier Detection in Human Disease Using the Paraclique Algorithm

Recent discoveries of distinct molecular subtypes have led to remarkable advances in treatment for a variety of diseases. While subtyping via unsupervised clustering has received a great deal of interest, most methods rely on basic statistical or machine learning methods. At the same time, techniques based on graph clustering, particularly clique-based strategies, have been successfully used to identify disease biomarkers and gene networks. A graph theoretical approach based on the paraclique algorithm is described that can easily be employed to identify putative disease subtypes and serve as an aid in outlier detection as well. The feasibility and potential effectiveness of this method is demonstrated on publicly available gene co-expression data derived from patient samples covering twelve different disease families.

Unsupervised dimensionality reduction for exposome research

Understanding the effect of the environment on human health has benefited from progress made in measuring the exposome. High resolution mass spectrometry (HRMS) has made it possible to measure small molecules across a large dynamic range, allowing researchers to study the role of low abundance environmental toxicants in causing human disease. HRMS data have a high dimensional structure (number of predictors >> number of observations), generating information on the abundance of many chemical features (predictors) which may be highly correlated. Unsupervised dimension reduction techniques can allow dimensionality reduction of the various features into components that capture the essence of the variability in the exposome dataset. We illustrate and discuss the relevance of three different unsupervised dimension reduction techniques: principal component analysis, factor analysis, and non-negative matrix factorization. We focus on the utility of each method in understanding the relationship between the exposome and a disease outcome and describe their strengths and limitations. While the utility of these methods is context specific, it remains important to focus on the interpretability of results from each method.

A robustness metric for biological data clustering algorithms

BACKGROUND

Cluster analysis is a core task in modern data-centric computation. Algorithmic choice is driven by factors such as data size and heterogeneity, the similarity measures employed, and the type of clusters sought. Familiarity and mere preference often play a significant role as well. Comparisons between clustering algorithms tend to focus on cluster quality. Such comparisons are complicated by the fact that algorithms often have multiple settings that can affect the clusters produced. Such a setting may represent, for example, a preset variable, a parameter of interest, or various sorts of initial assignments. A question of interest then is this: to what degree do the clusters produced vary as setting values change?

RESULTS

This work introduces a new metric, termed simply "robustness", designed to answer that question. Robustness is an easily-interpretable measure of the propensity of a clustering algorithm to maintain output coherence over a range of settings. The robustness of eleven popular clustering algorithms is evaluated over some two dozen publicly available mRNA expression microarray datasets. Given their straightforwardness and predictability, hierarchical methods generally exhibited the highest robustness on most datasets. Of the more complex strategies, the paraclique algorithm yielded consistently higher robustness than other algorithms tested, approaching and even surpassing hierarchical methods on several datasets. Other techniques exhibited mixed robustness, with no clear distinction between them.

CONCLUSIONS

Robustness provides a simple and intuitive measure of the stability and predictability of a clustering algorithm. It can be a useful tool to aid both in algorithm selection and in deciding how much effort to devote to parameter tuning.

Graph Theoretical Analysis of Genome-Scale Data: Examination of Gene Activation Occurring in the Setting of Community-Acquired Pneumonia

INTRODUCTION

We have previously reported evidence that Black individuals appear to have a significantly higher incidence of infection-related hospitalizations compared with White individuals. It is possible that the host immune response is responsible for this vital difference. In support of such a hypothesis, the aim of this study was to determine whether Black and White individuals exhibit differential whole blood gene network activation.

METHODS

We examined whole blood network activation in a subset of patients (n = 22 pairs, propensity score matched (1:1) Black and White patients) with community-acquired pneumonia (CAP) from the Genetic and Inflammatory Markers of Sepsis study. We employed day one whole blood transcriptomic data generated from this cohort and constructed co-expression graphs for each racial group. Pearson correlation coefficients were used to weight edges. Spectral thresholding was applied to ascribe significance. Innovative graph theoretical methods were then invoked to detect densely connected gene networks and provide differential structural analysis.

RESULTS

Propensity matching was employed to reduce potential bias due to confounding variables. Although Black and White patients had similar socio- and clinical demographics, we identified novel differences in molecular network activation-dense subgraphs known as paracliques that displayed complete gene connection for both White (three paracliques) and Black patients (one paraclique). Specifically, the genes that comprised the paracliques in the White patients include circadian loop, cell adhesion, mobility, proliferation, tumor suppression, NFκB, and chemokine signaling. However, the genes that comprised the paracliques in the Black patients include DNA and messenger RNA processes, and apoptosis signaling. We investigated the distribution of Black paracliques across White paracliques. Black patients had five paracliques (with almost complete connection) comprised of genes that are critical for host immune response widely distributed across 22 parcliques in the White population. Anchoring the analysis on two critical inflammatory mediators, interleukin (IL)-6 and IL-10 identified further differential network activation among the White and Black patient populations.

CONCLUSIONS

These results demonstrate that, at the molecular level, Black and White individuals may experience different activation patterns with CAP. Further validation of the gene networks we have identified may help pinpoint genetic factors that increase host susceptibility to community-acquired pneumonia, and may lay the groundwork for personalized management of CAP.

A multifactorial obesity model developed from nationwide public health exposome data and modern computational analyses

STATEMENT OF THE PROBLEM

Obesity is both multifactorial and multimodal, making it difficult to identify, unravel and distinguish causative and contributing factors. The lack of a clear model of aetiology hampers the design and evaluation of interventions to prevent and reduce obesity.

METHODS

Using modern graph-theoretical algorithms, we are able to coalesce and analyse thousands of inter-dependent variables and interpret their putative relationships to obesity. Our modelling is different from traditional approaches; we make no a priori assumptions about the population, and model instead based on the actual characteristics of a population. Paracliques, noise-resistant collections of highly-correlated variables, are differentially distilled from data taken over counties associated with low versus high obesity rates. Factor analysis is then applied and a model is developed.

RESULTS AND CONCLUSIONS

Latent variables concentrated around social deprivation, community infrastructure and climate, and especially heat stress were connected to obesity. Infrastructure, environment and community organisation differed in counties with low versus high obesity rates. Clear connections of community infrastructure with obesity in our results lead us to conclude that community level interventions are critical. This effort suggests that it might be useful to study and plan interventions around community organisation and structure, rather than just the individual, to combat the nation's obesity epidemic.

Robust Inference of Genetic Exchange Communities from Microbial Genomes Using TF-IDF

Bacteria and archaea can exchange genetic material across lineages through processes of lateral genetic transfer (LGT). Collectively, these exchange relationships can be modeled as a network and analyzed using concepts from graph theory. In particular, densely connected regions within an LGT network have been defined as genetic exchange communities (GECs). However, it has been problematic to construct networks in which edges solely represent LGT. Here we apply term frequency-inverse document frequency (TF-IDF), an alignment-free method originating from document analysis, to infer regions of lateral origin in bacterial genomes. We examine four empirical datasets of different size (number of genomes) and phyletic breadth, varying a key parameter (word length k) within bounds established in previous work. We map the inferred lateral regions to genes in recipient genomes, and construct networks in which the nodes are groups of genomes, and the edges natively represent LGT. We then extract maximum and maximal cliques (i.e., GECs) from these graphs, and identify nodes that belong to GECs across a wide range of k. Most surviving lateral transfer has happened within these GECs. Using Gene Ontology enrichment tests we demonstrate that biological processes associated with metabolism, regulation and transport are often over-represented among the genes affected by LGT within these communities. These enrichments are largely robust to change of k.

A novel approach to analyzing lung cancer mortality disparities: Using the exposome and a graph-theoretical toolchain

OBJECTIVES

The aim is to identify exposures associated with lung cancer mortality and mortality disparities by race and gender using an exposome database coupled to a graph theoretical toolchain.

METHODS

Graph theoretical algorithms were employed to extract paracliques from correlation graphs using associations between 2162 environmental exposures and lung cancer mortality rates in 2067 counties, with clique doubling applied to compute an absolute threshold of significance. Factor analysis and multiple linear regressions then were used to analyze differences in exposures associated with lung cancer mortality and mortality disparities by race and gender.

RESULTS

While cigarette consumption was highly correlated with rates of lung cancer mortality for both white men and women, previously unidentified novel exposures were more closely associated with lung cancer mortality and mortality disparities for blacks, particularly black women.

CONCLUSIONS

Exposures beyond smoking moderate lung cancer mortality and mortality disparities by race and gender.

POLICY IMPLICATIONS

An exposome approach and database coupled with scalable combinatorial analytics provides a powerful new approach for analyzing relationships between multiple environmental exposures, pathways and health outcomes. An assessment of multiple exposures is needed to appropriately translate research findings into environmental public health practice and policy.