Search

Modern biomedical data mining requires feature selection methods that can (1) be applied to large scale feature spaces (e.g. 'omics' data), (2) function in noisy problems, (3) detect complex patterns of association (e.g. gene-gene interactions), (4) be flexibly adapted to various problem domains and data types (e.g. genetic variants, gene expression, and clinical data) and (5) are computationally tractable. To that end, this work examines a set of filter-style feature selection algorithms inspired by the 'Relief' algorithm, i.e. Relief-Based algorithms (RBAs). We implement and expand these RBAs in an open source framework called ReBATE (Relief-Based Algorithm Training Environment). We apply a comprehensive genetic simulation study comparing existing RBAs, a proposed RBA called MultiSURF, and other established feature selection methods, over a variety of problems. The results of this study (1) support the assertion that RBAs are particularly flexible, efficient, and powerful feature selection methods that differentiate relevant features having univariate, multivariate, epistatic, or heterogeneous associations, (2) confirm the efficacy of expansions for classification vs. regression, discrete vs. continuous features, missing data, multiple classes, or class imbalance, (3) identify previously unknown limitations of specific RBAs, and (4) suggest that while MultiSURF∗ performs best for explicitly identifying pure 2-way interactions, MultiSURF yields the most reliable feature selection performance across a wide range of problem types. [ABSTRACT FROM AUTHOR]

Feature selection plays a critical role in biomedical data mining, driven by increasing feature dimensionality in target problems and growing interest in advanced but computationally expensive methodologies able to model complex associations. Specifically, there is a need for feature selection methods that are computationally efficient, yet sensitive to complex patterns of association, e.g. interactions, so that informative features are not mistakenly eliminated prior to downstream modeling. This paper focuses on Relief-based algorithms (RBAs), a unique family of filter-style feature selection algorithms that have gained appeal by striking an effective balance between these objectives while flexibly adapting to various data characteristics, e.g. classification vs. regression. First, this work broadly examines types of feature selection and defines RBAs within that context. Next, we introduce the original Relief algorithm and associated concepts, emphasizing the intuition behind how it works, how feature weights generated by the algorithm can be interpreted, and why it is sensitive to feature interactions without evaluating combinations of features. Lastly, we include an expansive review of RBA methodological research beyond Relief and its popular descendant, ReliefF. In particular, we characterize branches of RBA research, and provide comparative summaries of RBA algorithms including contributions, strategies, functionality, time complexity, adaptation to key data characteristics, and software availability. [ABSTRACT FROM AUTHOR]

Exposure to particulate matter ≤2.5 μm in diameter (PM 2.5) and ozone (O 3) has been linked to numerous harmful health outcomes. While epidemiologic evidence has suggested a positive association with type 2 diabetes (T2D), there is heterogeneity in findings. We evaluated exposures to PM 2.5 and O 3 across three large samples in the US using a harmonized approach for exposure assignment and covariate adjustment. Data were obtained from the Veterans Administration Diabetes Risk (VADR) cohort (electronic health records [EHRs]), the Reasons for Geographic and Racial Disparities in Stroke (REGARDS) cohort (primary data collection), and the Geisinger health system (EHRs), and reflect the years 2003–2016 (REGARDS) and 2008–2016 (VADR and Geisinger). New onset T2D was ascertained using EHR information on medication orders, laboratory results, and T2D diagnoses (VADR and Geisinger) or report of T2D medication or diagnosis and/or elevated blood glucose levels (REGARDS). Exposure was assigned using pollutant annual averages from the Downscaler model. Models stratified by community type (higher density urban, lower density urban, suburban/small town, or rural census tracts) evaluated likelihood of new onset T2D in each study sample in single- and two-pollutant models of PM 2.5 and O 3. In two pollutant models, associations of PM 2.5 , and new onset T2D were null in the REGARDS cohort except for in suburban/small town community types in models that also adjusted for NSEE, with an odds ratio (95% CI) of 1.51 (1.01, 2.25) per 5 μg/m3 of PM 2.5. Results in the Geisinger sample were null. VADR sample results evidenced nonlinear associations for both pollutants; the shape of the association was dependent on community type. Associations between PM 2.5 , O 3 and new onset T2D differed across three large study samples in the US. None of the results from any of the three study populations found strong and clear positive associations. • PM 2.5 and O 3 were minimally associated with new onset type 2 diabetes (T2D) in three large study samples in the United States. • Results did not substantially differ between single and two pollutant models. • Associations of PM 2.5 and O 3 with new onset T2D were impacted by adjustment for neighborhood socio-economic disadvantage. [ABSTRACT FROM AUTHOR]

Large-scale longitudinal studies evaluating influences of the built environment on risk for type 2 diabetes (T2D) are scarce, and findings have been inconsistent. To evaluate whether land use environment (LUE), a proxy of neighborhood walkability, is associated with T2D risk across different US community types, and to assess whether the association is modified by food environment. The Veteran's Administration Diabetes Risk (VADR) study is a retrospective cohort of diabetes-free US veteran patients enrolled in VA primary care facilities nationwide from January 1, 2008, to December 31, 2016, and followed longitudinally through December 31, 2018. A total of 4,096,629 patients had baseline addresses available in electronic health records that were geocoded and assigned a census tract-level LUE score. LUE scores were divided into quartiles, where a higher score indicated higher neighborhood walkability levels. New diagnoses for T2D were identified using a published computable phenotype. Adjusted time-to-event analyses using piecewise exponential models were fit within four strata of community types (higher-density urban, lower-density urban, suburban/small town, and rural). We also evaluated effect modification by tract-level food environment measures within each stratum. In adjusted analyses, higher LUE had a protective effect on T2D risk in rural and suburban/small town communities (linear quartile trend test p-value <0.001). However, in lower density urban communities, higher LUE increased T2D risk (linear quartile trend test p-value <0.001) and no association was found in higher density urban communities (linear quartile trend test p-value = 0.317). Particularly strong protective effects were observed for veterans living in suburban/small towns with more supermarkets and more walkable spaces (p-interaction = 0.001). Among veterans, LUE may influence T2D risk, particularly in rural and suburban communities. Food environment may modify the association between LUE and T2D. [ABSTRACT FROM AUTHOR]