Non-random sampling leads to biased estimates of transcriptome association.

Integration of independent data resources across -omics platforms offers transformative opportunity for novel clinical and biological discoveries. However, application of emerging analytic methods in the context of selection bias represents a noteworthy and pervasive challenge. We hypothesize that combining differentially selected samples for integrated transcriptome analysis will lead to bias in the estimated association between predicted expression and the trait. Our results are based on in silico investigations and a case example focused on body mass index across four well-described cohorts apparently derived from markedly different populations. Our findings suggest that integrative analysis can lead to substantial relative bias in the estimate of association between predicted expression and the trait. The average estimate of association ranged from 51.3% less than to 96.7% greater than the true value for the biased sampling scenarios considered, while the average error was - 2.7% for the unbiased scenario. The corresponding 95% confidence interval coverage rate ranged from 46.4% to 69.5% under biased sampling, and was equal to 75% for the unbiased scenario. Inverse probability weighting with observed and estimated weights is applied as one corrective measure and appears to reduce the bias and improve coverage. These results highlight a critical need to address selection bias in integrative analysis and to use caution in interpreting findings in the presence of different sampling mechanisms between groups.

HIV-1-negative female sex workers sustain high cervical IFNɛ, low immune activation, and low expression of HIV-1-required host genes.

Sex workers practicing in high HIV endemic areas have been extensively targeted to test anti-HIV prophylactic strategies. We hypothesize that in women with high levels of genital exposure to semen changes in cervico-vaginal mucosal and/or systemic immune activation will contribute to a decreased susceptibility to HIV-1 infection. To address this question, we assessed sexual activity and immune activation status (in peripheral blood), as well as cellular infiltrates and gene expression in ectocervical mucosa biopsies in female sex workers (FSWs; n=50), as compared with control women (CG; n=32). FSWs had low-to-absent HIV-1-specific immune responses with significantly lower CD38 expression on circulating CD4(+) or CD8(+) T-cells (both: P<0.001) together with lower cervical gene expression of genes associated with leukocyte homing and chemotaxis. FSWs also had increased levels of interferon-ɛ (IFNɛ) gene and protein expression in the cervical epithelium together with reduced expression of genes associated with HIV-1 integration and replication. A correlative relationship between semen exposure and elevated type-1 IFN expression in FSWs was also established. Overall, our data suggest that long-term condomless sex work can result in multiple changes within the cervico-vaginal compartment that would contribute to sustaining a lower susceptibility for HIV-1 infection in the absence of HIV-specific responses.

Prediction based classification for longitudinal biomarkers.

Assessment of circulating CD4 count change over time in HIV-infected subjects on antiretroviral therapy (ART) is a central component of disease monitoring. The increasing number of HIV-infected subjects starting therapy and the limited capacity to support CD4 count testing within resource-limited settings have fueled interest in identifying correlates of CD4 count change such as total lymphocyte count, among others. The application of modeling techniques will be essential to this endeavor due to the typically non-linear CD4 trajectory over time and the multiple input variables necessary for capturing CD4 variability. We propose a prediction based classification approach that involves first stage modeling and subsequent classification based on clinically meaningful thresholds. This approach draws on existing analytical methods described in the receiver operating characteristic curve literature while presenting an extension for handling a continuous outcome. Application of this method to an independent test sample results in greater than 98% positive predictive value for CD4 count change. The prediction algorithm is derived based on a cohort of n = 270 HIV-1 infected individuals from the Royal Free Hospital, London who were followed for up to three years from initiation of ART. A test sample comprised of n = 72 individuals from Philadelphia and followed for a similar length of time is used for validation. Results suggest that this approach may be a useful tool for prioritizing limited laboratory resources for CD4 testing after subjects start antiretroviral therapy.

Estimating and testing haplotype-trait associations in non-diploid populations.

Malaria is an infectious disease that is caused by a group of parasites of the genus Plasmodium. Characterizing the association between polymorphisms in the parasite genome and measured traits in an infected human host may provide insight into disease aetiology and ultimately inform new strategies for improved treatment and prevention. This, however, presents an analytic challenge since individuals are often multiply infected with a variable and unknown number of genetically diverse parasitic strains. In addition, data on the alignment of nucleotides on a single chromosome, which is commonly referred to as haplotypic phase, is not generally observed. An expectation-maximization algorithm for estimating and testing associations between haplotypes and quantitative traits has been described for diploid (human) populations. We extend this method to account for both the uncertainty in haplotypic phase and the variable and unknown number of infections in the malaria setting. Further extensions are described for the human immunodeficiency virus quasi-species setting. A simulation study is presented to characterize performance of the method. Application of this approach to data arising from a cross-sectional study of n=126 multiply infected children in Uganda reveals some interesting associations requiring further investigation.

Tree-Based Methods for Discovery of Association between Flow Cytometry Data and Clinical Endpoints.

We demonstrate the application and comparative interpretations of three tree-based algorithms for the analysis of data arising from flow cytometry: classification and regression trees (CARTs), random forests (RFs), and logic regression (LR). Specifically, we consider the question of what best predicts CD4 T-cell recovery in HIV-1 infected persons starting antiretroviral therapy with CD4 count between 200 and 350 cell/muL. A comparison to a more standard contingency table analysis is provided. While contingency table analysis and RFs provide information on the importance of each potential predictor variable, CART and LR offer additional insight into the combinations of variables that together are predictive of the outcome. In all cases considered, baseline CD3-DR-CD56+CD16+ emerges as an important predictor variable, while the tree-based approaches identify additional variables as potentially informative. Application of tree-based methods to our data suggests that a combination of baseline immune activation states, with emphasis on CD8 T-cell activation, may be a better predictor than any single T-cell/innate cell subset analyzed. Taken together, we show that tree-based methods can be successfully applied to flow cytometry data to better inform and discover associations that may not emerge in the context of a univariate analysis.

Mixed modeling and multiple imputation for unobservable genotype clusters.

Understanding the genetic contributions to complex diseases will require consideration of interaction across multiple genes and environmental factors. At the same time, capturing information on allelic phase, that is, whether alleles within a gene are in cis (on the same chromosome) or in trans (on different chromosomes), is critical when using haplotypic approaches in disease association studies. This paper proposes a combination of mixed modeling and multiple imputation for assessing high-order genotype-phenotype associations while accounting for the uncertainty in phase inherent in population-based association studies. This method provides a flexible statistical framework for controlling for potential confounders and assessing gene-environment and gene-gene interactions in studies of unrelated individuals where the haplotypic phase is generally unobservable. The proposed method is applied to a cohort of 626 subjects with human immunodeficiency virus (HIV) to assess the potential contribution of four genes, apolipoprotein-C-III, apolipoprotein-E, endothelial lipase and hepatic lipase in predicting lipid abnormalities. A simulation study is also presented to describe the method performance.

Mixed modelling to characterize genotype-phenotype associations.

We propose using mixed effects models to characterize the association between multiple gene polymorphisms, environmental factors and measures of disease progression. Characterizing high-order gene-gene and gene-environment interactions presents an analytic challenge due to the large number of candidate genes and the complex, undescribed interactions among them. Several approaches have been proposed recently to reduce the number of candidate genes and post hoc approaches to identify gene-gene interactions are described. However, these approaches may be inadequate for identifying high-order interactions in the absence of main effects and generally do not permit us to control for potential confounders. We describe how mixed effects models and related testing procedures overcome these limitations and apply this approach to data from a cohort of subjects at risk for cardiovascular disease. Four (4) genetic polymorphisms in three genes of the same gene family are considered. The proposed modelling approach allows us first to test whether there is a significant genetic contribution to the variability observed in our disease outcome. This contribution may be through main effects of multi-locus genotypes or through an interaction between genotype and environmental factors. This approach also enables us to identify specific multi-locus genotypes that interact with environmental factors in predicting the outcome. Mixed effects models provide a flexible statistical framework for controlling for potential confounders and identifying interactions among multiple genes and environmental factors that explain the variability in measures of disease progression.

MDR and PRP: a comparison of methods for high-order genotype-phenotype associations.

Complex diseases such as cardiovascular disease are likely due to the effects of high-order interactions among multiple genes and demographic factors. Therefore, in order to understand their underlying biological mechanisms, we need to consider simultaneously the effects of genotypes across multiple loci. Statistical methods such as multifactor dimensionality reduction (MDR), the combinatorial partitioning method (CPM), recursive partitioning (RP), and patterning and recursive partitioning (PRP) are designed to uncover complex relationships without relying on a specific model for the interaction, and are therefore well-suited to this data setting. However, the theoretical overlap among these methods and their relative merits have not been well characterized. In this paper we demonstrate mathematically that MDR is a special case of RP in which (1) patterns are used as predictors (PRP), (2) tree growth is restricted to a single split, and (3) misclassification error is used as the measure of impurity. Both approaches are applied to a case-control study assessing the effect of eleven single nucleotide polymorphisms on coronary artery calcification in people at risk for cardiovascular disease.

Characterizing classes of antiretroviral drugs by genotype.

This paper develops methods for using HIV-1 genotypic information to group patients who are expected to have similar patterns of sensitivity or resistance to two or more drugs. The methods presented are an extension of prediction based classification to handle multiple drug responses. Here, the goal is to determine the probability that one antiretroviral therapy will be more favourable than another for an individual given the specific genotypic or other characteristics of the infecting viral population. This approach requires a model relating genotype to a vector of drug specific phenotypic responses. A comparison of Nelfinavir and Indinavir is provided using 2746 protease sequences and corresponding in vitro sensitivity assays provided to us by the Virco Group.

Characterizing the relationship between HIV-1 genotype and phenotype: prediction-based classification.

This paper establishes a framework for understanding the complex relationships between HIV-1 genotypic markers of resistance to antiretroviral drugs and clinical measures of disease progression. A new classification scheme based on the probabilities of how new patients will respond to antiretroviral therapy given the available data is proposed as a method for distinguishing among groups of viral sequences. This approach draws from existing cluster analysis, discriminant analysis, and recursive partitioning techniques and requires a model relating genotypic characteristics to phenotypic response. A data set of 2,746 sequences and the corresponding Indinavir 50% inhibitory concentrations are described and used for illustrative purposes.

Methods for investigation of the relationship between drug-susceptibility phenotype and human immunodeficiency virus type 1 genotype with applications to AIDS clinical trials group 333.

Use of human immunodeficiency virus (HIV) drug-resistance testing in therapeutic decision making may be aided by understanding the relationship between results of genotypic and drug-susceptibility phenotypic assays. We investigated this relationship by applying 3 different statistical methods-cluster analysis, recursive partitioning, and linear discriminant analysis-to results for 72 patients followed in the Adult AIDS Clinical Trials Group (ACTG) protocol 333. ACTG 333 was a multicenter, randomized trial comparing 2 formulations of saquinavir (SQV) to indinavir (IDV) in patients with extensive hard-gel SQV experience. Data include protease amino acid sequences and 50% inhibitory concentrations for SQV and IDV at baseline. The 3 methods give similar results showing the association of mutations at codons 10, 63, 71, and 90 with in vitro resistance to IDV and SQV. Recursive partitioning is especially useful because it can identify interactions among mutations at different codons and accommodates many types of data as well as missing observations.

A murine model of myocardial microvascular thrombosis.

Disorders of hemostasis lead to vascular pathology. Endothelium-derived gene products play a critical role in the formation and degradation of fibrin. We sought to characterize the importance of these locally produced factors in the formation of fibrin in the cardiac macrovasculature and microvasculature. This study used mice with modifications of the thrombomodulin (TM) gene, the tissue-type plasminogen activator (tPA) gene, and the urokinase-type plasminogen activator (uPA) gene. The results revealed that tPA played the most important role in local regulation of fibrin deposition in the heart, with lesser contributions by TM and uPA (least significant). Moreover, a synergistic relationship in fibrin formation existed in mice with concomitant modifications of tPA and TM, resulting in myocardial necrosis and depressed cardiac function. The data were fit to a statistical model that may offer a foundation for examination of hemostasis-regulating gene interactions.