A comparison of analytical methods for genetic association studies.

The explosion of genetic information over the last decade presents an analytical challenge for genetic association studies. As the number of genetic variables examined per individual increases, both variable selection and statistical modeling tasks must be performed during analysis. While these tasks could be performed separately, coupling them is necessary to select meaningful variables that effectively model the data. This challenge is heightened due to the complex nature of the phenotypes under study and the complex underlying genetic etiologies. To address this problem, a number of novel methods have been developed. In the current study, we compare the performance of six analytical approaches to detect both main effects and gene-gene interactions in a range of genetic models. Multifactor dimensionality reduction, grammatical evolution neural networks, random forests, focused interaction testing framework, step-wise logistic regression, and explicit logistic regression were compared. As one might expect, the relative success of each method is context dependent. This study demonstrates the strengths and weaknesses of each method and illustrates the importance of continued methods development.

Household food insecurity associated with antiretroviral therapy adherence among HIV-infected patients in Windhoek, Namibia.

OBJECTIVE: Food insecurity is emerging as an important barrier to antiretroviral therapy (ART) adherence. The objective of this study was to determine if food insecurity is associated with poor ART adherence among HIV-positive adults in a resource-limited setting that uses the public health model of delivery. DESIGN: A cross-sectional study using a 1-time questionnaire and routinely collected pharmacy data. METHODS: Participants were HIV-infected adults on ART at the public ART clinics in Windhoek, Namibia: Katutura State Hospital, Katutura Health Centre, and Windhoek Central Hospital. Food insecurity was measured by the Household Food Insecurity Access Scale (HFIAS). Adherence was assessed by the pharmacy adherence measure medication possession ratio (MPR). Multivariate regression was used to assess whether food insecurity was associated with ART adherence. RESULTS: Among 390 participants, 7% were food secure, 25% were mildly or moderately food insecure and 67% were severely food insecure. In adjusted analyses, severe household food insecurity was associated with MPR <80% [odds ratio (OR), 3.84; 95% confidence interval (CI): 1.65 to 8.95]. Higher household health care spending (OR, 1.92; 95% CI, 1.02 to 3.57) and longer duration of ART (OR, 0.82; 95% CI: 0.70 to 0.97) were also associated with <80% MPR. CONCLUSIONS: Severe household food insecurity is present in more than half of the HIV-positive adults attending a public ART clinic in Windhoek, Namibia and is associated with poor ART adherence as measured by MPR. Ensuring reliable access to food should be an important component of ART delivery in resource-limited settings using the public health model of care.

Hormone therapy, estrogen metabolism, and risk of breast cancer in the Women's Health Initiative Hormone Therapy Trial.

BACKGROUND: In the Women's Health Initiative Hormone Trials (WHI-HT), breast cancer risk was increased with estrogen plus progestin (E+P) but not with unopposed estrogen (E-alone). We hypothesized that E+P would preferentially metabolize to 16α-hydroxyestrone (16α-OHE1) rather than 2-hydroxyestrone (2-OHE1), and that breast cancer risk would be associated with baseline and 1 year changes in estrogen metabolites: positively for 16α-OHE1 levels and negatively for levels of 2-OHE-1 and the 2:16 ratio. METHODS: In a prospective case-control study nested in the WHI-HT, 845 confirmed breast cancer cases were matched to 1,690 controls by age and ethnicity. Using stored serum, 2-OHE1 and 16α-OHE1 levels were measured by enzyme immunoassay at baseline, and for those randomized to active treatment (n = 1,259), at 1 year. RESULTS: The 1-year increase in 16α-OHE1 was greater with E+P than E-alone (median 55.5 pg/mL vs. 43.5 pg/mL, P < 0.001), but both increased 2-OHE1 by ∼300 pg/mL. Breast cancer risk was modestly associated with higher baseline levels of 2-OHE1 and the 2:16 ratio, and for estrogen receptor+/progesterone+ cases only, higher baseline 16α-OHE1 levels. For those randomized to active treatment, breast cancer risk was associated with greater increase in 2-OHE-1 and the 2:16 ratio, but associations were not significant. CONCLUSIONS: Although E+P modestly increased 16α-OHE1 more than E-alone, increase in 16α-OHE1 was not associated with breast cancer. IMPACT: Study results do not explain differences between the WHI E+P and WHI E-alone breast cancer results but metabolism of oral HT, which may explain smaller than expected increase in breast cancer compared with endogenous estrogens.

Power of grammatical evolution neural networks to detect gene-gene interactions in the presence of error.

BACKGROUND: With the advent of increasingly efficient means to obtain genetic information, a great insurgence of data has resulted, leading to the need for methods for analyzing this data beyond that of traditional parametric statistical approaches. Recently we introduced Grammatical Evolution Neural Network (GENN), a machine-learning approach to detect gene-gene or gene-environment interactions, also known as epistasis, in high dimensional genetic epidemiological data. GENN has been shown to be highly successful in a range of simulated data, but the impact of error common to real data is unknown. In the current study, we examine the power of GENN to detect interesting interactions in the presence of noise due to genotyping error, missing data, phenocopy, and genetic heterogeneity. Additionally, we compare the performance of GENN to that of another computational method - Multifactor Dimensionality Reduction (MDR). FINDINGS: GENN is extremely robust to missing data and genotyping error. Phenocopy in a dataset reduces the power of both GENN and MDR. GENN is reasonably robust to genetic heterogeneity and find that in some cases GENN has substantially higher power than MDR to detect functional loci in the presence of genetic heterogeneity. CONCLUSION: GENN is a promising method to detect gene-gene interaction, even in the presence of common types of error found in real data.

A Balanced Accuracy Fitness Function Leads to Robust Analysis using Grammatical Evolution Neural Networks in the Case of Class Imbalance.

Grammatical Evolution Neural Networks (GENN) is a computational method designed to detect gene-gene interactions in genetic epidemiology, but has so far only been evaluated in situations with balanced numbers of cases and controls. Real data, however, rarely has such perfectly balanced classes. In the current study, we test the power of GENN to detect interactions in data with a range of class imbalance using two fitness functions (classification error and balanced error), as well as data re-sampling. We show that when using classification error, class imbalance greatly decreases the power of GENN. Re-sampling methods demonstrated improved power, but using balanced accuracy resulted in the highest power. Based on the results of this study, balanced error has replaced classification error in the GENN algorithm.

Linkage Disequilibrium in Genetic Association Studies Improves the Performance of Grammatical Evolution Neural Networks.

One of the most important goals in genetic epidemiology is the identification of genetic factors/features that predict complex diseases. The ubiquitous nature of gene-gene interactions in the underlying etiology of common diseases creates an important analytical challenge, spurring the introduction of novel, computational approaches. One such method is a grammatical evolution neural network (GENN) approach. GENN has been shown to have high power to detect such interactions in simulation studies, but previous studies have ignored an important feature of most genetic data: linkage disequilibrium (LD). LD describes the non-random association of alleles not necessarily on the same chromosome. This results in strong correlation between variables in a dataset, which can complicate analysis. In the current study, data simulations with a range of LD patterns are used to assess the impact of such correlated variables on the performance of GENN. Our results show that not only do patterns of strong LD not decrease the power of GENN to detect genetic associations, they actually increase its power.