[Molenberghs, Geert] Univ Hasselt, Hasselt, Belgium. [Molenberghs, Geert] Katholieke Univ Leuven, BioStat 1, Leuven, Belgium.

Our increased understanding of the mechanistic heterogeneity of diseases has pushed the development of targeted therapeutics. We do not expect all patients with a given disease to benefit from a targeted drug; only those in the target population. That is, those with sufficient dysregulation in the biomolecular pathway targeted by treatment. However, due to complexity of the pathway, and/or technical issues with our characterizing assay, it is often hard to characterize the target population until well into large-scale clinical trials. This has stimulated the development of adaptive enrichment trials; clinical trials in which the target population is adaptively learned; and enrollment criteria are adaptively updated to reflect this growing understanding. This paper proposes a framework for group-sequential adaptive enrichment trials. Building on the work of Simon & Simon (2013). Adaptive enrichment designs for clinical trials. Biostatistics 14(4), 613-625), it includes a frequentist hypothesis test at the end of the trial. However, it uses Bayesian methods to optimize the decisions required during the trial (regarding how to restrict enrollment) and Bayesian methods to estimate effect size, and characterize the target population at the end of the trial. This joint frequentist/Bayesian design combines the power of Bayesian methods for decision making with the use of a formal hypothesis test at the end of the trial to preserve the studywise probability of a type I error.

This article proposes a procedure for describing the relationship between high-dimensional data sets, such as multimodal brain images and genetic data. We propose a supervised technique to incorporate the clinical outcome to determine a score, which is a linear combination of variables with hieratical structures to multimodalities. This approach is expected to obtain interpretable and predictive scores. The proposed method was applied to a study of Alzheimer's disease (AD). We propose a diagnostic method for AD that involves using whole-brain magnetic resonance imaging (MRI) and positron emission tomography (PET), and we select effective brain regions for the diagnostic probability and investigate the genome-wide association with the regions using single nucleotide polymorphisms (SNPs). The two-step dimension reduction method, which we previously introduced, was considered applicable to such a study and allows us to partially incorporate the proposed method. We show that the proposed method offers classification functions with feasibility and reasonable prediction accuracy based on the receiver operating characteristic (ROC) analysis and reasonable regions of the brain and genomes. Our simulation study based on the synthetic structured data set showed that the proposed method outperformed the original method and provided the characteristic for the supervised feature.

RNA-Seq data characteristically exhibits large variances, which need to be appropriately accounted for in any proposed model. We first explore the effects of this variability on the maximum likelihood estimator (MLE) of the dispersion parameter of the negative binomial distribution, and propose instead to use an estimator obtained via maximization of the marginal likelihood in a conjugate Bayesian framework. We show, via simulation studies, that the marginal MLE can better control this variation and produce a more stable and reliable estimator. We then formulate a conjugate Bayesian hierarchical model, and use this new estimator to propose a Bayesian hypothesis test to detect differentially expressed genes in RNA-Seq data. We use numerical studies to show that our much simpler approach is competitive with other negative binomial based procedures, and we use a real data set to illustrate the implementation and flexibility of the procedure.

The area under the curve (AUC) statistic is a common measure of model performance in a binary regression model. Nested models are used to ascertain whether the AUC statistic increases when new factors enter the model. The regression coefficient estimates used in the AUC statistics are computed using the maximum rank correlation methodology. Typically, inference for the difference in AUC statistics from nested models is derived under asymptotic normality. In this work, it is demonstrated that the asymptotic normality is true only when at least one of the new factors is associated with the binary outcome. When none of the new factors are associated with the binary outcome, the asymptotic distribution for the difference in AUC statistics is a linear combination of chi-square random variables. Further, when at least one new factor is associated with the outcome and the population difference is small, a variance stabilizing reparameterization improves the asymptotic normality of the AUC difference statistic. A confidence interval using this reparameterization is developed and simulations are generated to determine their coverage properties. The derived confidence interval provides information on the magnitude of the added value of new factors and enables investigators to weigh the size of the improvement against potential costs associated with the new factors. A pancreatic cancer data example is used to illustrate this approach.

A large number of cancer drugs have been developed to target particular genes/pathways that are crucial for cancer growth. Drugs that share a molecular target may also have some common predictive omic features, e.g., somatic mutations or gene expression. Therefore, it is desirable to analyze these drugs as a group to identify the associated omic features, which may provide biological insights into the underlying drug response. Furthermore, these omic features may be robust predictors for any drug sharing the same target. The high dimensionality and the strong correlations among the omic features are the main challenges of this task. Motivated by this problem, we develop a new method for high-dimensional bilevel feature selection using a group of response variables that may share a common set of predictors in addition to their individual predictors. Simulation results show that our method has a substantially higher sensitivity and specificity than existing methods. We apply our method to two large-scale drug sensitivity studies in cancer cell lines. Both within-study and between-study validation demonstrate the good efficacy of our method.

In multidimensional cancer omics studies, one subject is profiled on multiple layers of omics activities. In this article, the goal is to integrate multiple types of omics measurements, identify markers, and build a model for cancer outcome. The proposed analysis is achieved in two steps. In the first step, we analyze the regulation among different types of omics measurements, through the construction of linear regulatory modules (LRMs). The LRMs have sound biological basis, and their construction differs from the existing analyses by modeling the regulation of sets of gene expressions (GEs) by sets of regulators. The construction is realized with the assistance of regularized singular value decomposition. In the second step, the proposed cancer outcome model includes the regulated GEs, "residuals" of GEs, and "residuals" of regulators, and we use regularized estimation to select relevant markers. Simulation shows that the proposed method outperforms the alternatives with more accurate marker identification. We analyze the The Cancer Genome Atlas data on cutaneous melanoma and lung adenocarcinoma and obtain meaningful results.