A roadmap of clustering algorithms: finding a match for a biomedical application.

Clustering is ubiquitously applied in bioinformatics with hierarchical clustering and k-means partitioning being the most popular methods. Numerous improvements of these two clustering methods have been introduced, as well as completely different approaches such as grid-based, density-based and model-based clustering. For improved bioinformatics analysis of data, it is important to match clusterings to the requirements of a biomedical application. In this article, we present a set of desirable clustering features that are used as evaluation criteria for clustering algorithms. We review 40 different clustering algorithms of all approaches and datatypes. We compare algorithms on the basis of desirable clustering features, and outline algorithms' benefits and drawbacks as a basis for matching them to biomedical applications.

Epidemiology of physical and mental comorbidity in Canada and implications for health-related quality of life, suicidal ideation, and healthcare utilization: A nationwide cross-sectional study.

BACKGROUND: The substantial burden of physical and mental comorbidity is increasingly gaining attention, but a comprehensive evaluation of this is limited in Canada. This study aimed to investigate the prevalence of physical and mental comorbidity and its implications in Canada. METHODS: We used nationally representative data from Canadian Community Health Survey, 2014. We included individuals who were aged ≥18 years and excluded those who had missing information on physical or mental disorders. Chronic diseases referred to both physical and mental disorders. RESULTS: Respondents included in our analysis represented 27,221,856 Canadians aged ≥18 years. Of these, 53.9% (95% CI 53.1-54.6) had one or more chronic diseases, 11.5% (95% CI 11.0-12.0) had mental disorder, and 8.4% (95% CI 8.0-8.8) had physical and mental comorbidity. Compared with those without chronic diseases, people with one or more chronic diseases had higher sex- and age-adjusted prevalence of severe impairment of health-related quality of life (HRQoL), suicidal ideation, and healthcare utilization; and the risks increased consistently with the number of chronic diseases. However, among those with the same number of chronic diseases, people with mental disorder or physical and mental comorbidity were more likely to have these adverse consequences than people with only physical disorders. LIMITATIONS: Our study was based on self-reported data, and included only major chronic diseases rather than all probable chronic diseases. CONCLUSIONS: Physical and mental comorbidity is prevalent in Canada and should be addressed with appropriate interventions considering its excessive adverse impact on HRQoL, suicidal ideation and healthcare utilization.

Associations of sleep problems with health-risk behaviors and psychological well-being among Canadian adults.

OBJECTIVES: Examine the associations of sleep problems with health-risk behaviors and psychological well-being in a representative sample of Canadian adults. DESIGN: Cross-sectional. SETTING: The 2011-2012 Canadian Community Health Survey (CCHS, conducted by Statistics Canada). PARTICIPANTS: Of all individuals taking part in the 2011-2012 CCHS, 42,600 participants aged ≥18 years from five provinces/territories (Nova Scotia, Quebec, Manitoba, Alberta, and Yukon) who participated in the sleep survey module were selected for this study. MEASUREMENTS: Health conditions were self-reported. Sleep problems referred to extreme sleep durations (either <5 or ≥10 hours) and insomnia symptom. Health-risk behaviors included physical inactivity, daily smoking, highly sedentary behavior, and insufficient fruit and vegetable consumption. Worse psychological well-being included having worse self-rated general health, worse self-rated mental health, and worse sense of belonging, and being dissatisfied with life. RESULTS: The participants represented 10,614,600 Canadian adults aged ≥18 years from the five abovementioned provinces/territories. A significantly higher prevalence of all health-risk behaviors and worse psychological well-being was found among participants with extreme sleep durations (than those with 7 to <8 hours) and insomnia symptom (than those without insomnia symptom). After multivariate adjustment, extreme sleep durations and insomnia symptom were still independently associated with increased odds of all health-risk behaviors and worse psychological well-being. CONCLUSIONS: Both extreme sleep durations and insomnia symptom were independently associated with health-risk behaviors and worse psychological well-being among Canadian adults.

Efficient layered density-based clustering of categorical data.

A challenge involved in applying density-based clustering to categorical biomedical data is that the "cube" of attribute values has no ordering defined, making the search for dense subspaces slow. We propose the HIERDENC algorithm for hierarchical density-based clustering of categorical data, and a complementary index for searching for dense subspaces efficiently. The HIERDENC index is updated when new objects are introduced, such that clustering does not need to be repeated on all objects. The updating and cluster retrieval are efficient. Comparisons with several other clustering algorithms showed that on large datasets HIERDENC achieved better runtime scalability on the number of objects, as well as cluster quality. By fast collapsing the bicliques in large networks we achieved an edge reduction of as much as 86.5%. HIERDENC is suitable for large and quickly growing datasets, since it is independent of object ordering, does not require re-clustering when new data emerges, and requires no user-specified input parameters.

Medical text analytics tools for search and classification.

A text-analytic tool has been developed that accepts clinical medical data as input in order to produce patient details. The integrated tool has the following four characteristics. 1) It has a graphical user interface. 2) It has a free-text search tool that is designed to retrieve records using keywords such as "MI" for myocardial infarction. The result set is a display of those sentences in the medical records that contain the keywords. 3) It has three tools to classify patients based on the likelihood of being diagnosed for myocardial infarction, hypertension, or their smoking status. 4) A summary is generated for each patient selected. Large medical data sets provided by the Institute for Clinical Evaluative Sciences were used during the project.

Association between sleep problems and health-related quality of life in Canadian adults with chronic diseases.

OBJECTIVES: This study aimed to explore the association between sleep problems and health-related quality of life (HRQoL) in Canadian adults with chronic diseases, and whether mental illness can mediate the association. METHODS: Data were drawn from the Canadian Community Health Survey, 2015. A total of 10,900 participants aged ≥18 years and diagnosed with chronic diseases were enrolled in this study. RESULTS: Of these participants, 23.6% (95% CI 22.1, 25.2) suffered from severe impairment of HRQoL. Extreme sleep durations, including both short (<5, 5 to <6, and 6 to <7 h) and long (9 to <10, and ≥10 h) sleep durations, were significantly associated with severe impairment of HRQoL (compared to 7 to <8 h). Insomnia was also independently associated with severe impairment of HRQoL when compared to those without insomnia. In the mediation analyses, mental illness was shown to partly mediate the associations of extreme sleep durations and insomnia with severe impairment of HRQoL. CONCLUSIONS: In conclusion, both extreme sleep durations and insomnia were independently associated with severe impairment of HRQoL in adults with chronic diseases, and mental illness partly mediated the association.

Clustering by common friends finds locally significant proteins mediating modules.

MOTIVATION: Much research has been dedicated to large-scale protein interaction networks including the analysis of scale-free topologies, network modules and the relation of domain-domain to protein-protein interaction networks. Identifying locally significant proteins that mediate the function of modules is still an open problem. METHOD: We use a layered clustering algorithm for interaction networks, which groups proteins by the similarity of their direct neighborhoods. We identify locally significant proteins, called mediators, which link different clusters. We apply the algorithm to a yeast network. RESULTS: Clusters and mediators are organized in hierarchies, where clusters are mediated by and act as mediators for other clusters. We compare the clusters and mediators to known yeast complexes and find agreement with precision of 71% and recall of 61%. We analyzed the functions, processes and locations of mediators and clusters. We found that 55% of mediators to a cluster are enriched with a set of diverse processes and locations, often related to translocation of biomolecules. Additionally, 82% of clusters are enriched with one or more functions. The important role of mediators is further corroborated by a comparatively higher degree of conservation across genomes. We illustrate the above findings with an example of membrane protein translocation from the cytoplasm to the inner nuclear membrane. AVAILABILITY: All software is freely available under Supplementary information.

Mining significant high utility gene regulation sequential patterns.

BACKGROUND: Mining frequent gene regulation sequential patterns in time course microarray datasets is an important mining task in bioinformatics. Although finding such patterns are of paramount important for studying a disease, most existing work do not consider gene-disease association during gene regulation sequential pattern discovery. Moreover, they consider more absent/existence effects of genes during the mining process than taking the degrees of genes expression into account. Consequently, such techniques discover too many patterns which may not represent important information to biologists to investigate the relationships between the disease and underlying reasons hidden in gene regulation sequences. RESULTS: We propose a utility model by considering both the gene-disease association score and their degrees of expression levels under a biological investigation. We propose an efficient method called Top-HUGS, for discoverying significant high utility gene regulation sequential patterns from a time-course microarray dataset. CONCLUSIONS: In this study, the proposed methods were evaluated on a publicly available time course microarray dataset. The experimental results show higher accuracies compared to the baseline methods. Our proposed methods found that several new gene regulation sequential patterns involved in such patterns were useful for biologists and provided further insights into the mechanisms underpinning biological processes. To effectively work with the proposed method, a web interface is developed to our system using Java. To the best of our knowledge, this is the first demonstration for significant high utility gene regulation sequential pattern discovery.

Conceptualizing type 2 diabetes and its management.

Type 2 diabetes is growing worldwide due to population growth, increased rates of obesity, unhealthy diet, and physical inactivity. Risk assessment methods can effectively evaluate the risk of diabetes, and a healthy lifestyle can significantly reduce risk or prevent complications of type 2 diabetes. However, risk assessment alone has not significantly improved poor adherence to recommended medical interventions and lifestyle changes. This paper focuses on the challenge of nonadherence and posits that improving adherence requires tailoring interventions that explicitly consider the social determinants of health.

Finding molecular complexes through multiple layer clustering of protein interaction networks.

Clustering protein-protein interaction networks (PINs) helps to identify complexes that guide the cell machinery. Clustering algorithms often create a flat clustering, without considering the layered structure of PINs. We propose the MULIC clustering algorithm that produces layered clusters. We applied MULIC to five PINs. Clusters correlate with known MIPS protein complexes. For example, a cluster of 79 proteins overlaps with a known complex of 88 proteins. Proteins in top cluster layers tend to be more representative of complexes than proteins in bottom layers. Lab work on finding unknown complexes or determining drug effects can be guided by top layer proteins.

Bi-level clustering of mixed categorical and numerical biomedical data.

Biomedical data sets often have mixed categorical and numerical types, where the former represent semantic information on the objects and the latter represent experimental results. We present the BILCOM algorithm for 'Bi-Level Clustering of Mixed categorical and numerical data types'. BILCOM performs a pseudo-Bayesian process, where the prior is categorical clustering. BILCOM partitions biomedical data sets of mixed types, such as hepatitis, thyroid disease and yeast gene expression data with Gene Ontology annotations, more accurately than if using one type alone.