This proposal is a comprehensive research plan for establishing a general framework for measuring available statistical information in gene mapping studies. The key methodological challenge is to find a measure that (1) is a reliable index of the relative information specific to the purpose of a study, (2) conditions on particular data sets, (3) is robust in the sense of general applicability, including to small data sets, (4) is easy to compute, and (5) is subject to sensible combination axioms. Dealing with all these criteria simultaneously requires a careful combination of Bayesian and frequentist methods, especially for small samples. The PIs propose to investigate a large-sample framework involving likelihood functions only, and a small-sample framework from a robust Bayesian perspective. The robust Bayesian approach takes full advantage of the Bayesian formulation in deriving information measures with desirable coherence properties, and at the same time it seeks measures that are robust to various specifications and thus are more generally applicable. The PIs also propose to investigate several specific measures at two levels. At the more general level, the PIs will study and compare these measures in terms of their general behaviors and applicability, which are not restricted to the genetic setting. At the more specific level, the PIs plan to evaluate and apply these measures in specific genetic applications, including allele-sharing methods, methods for fine-scale genetic mapping (e.g., haplotype-sharing methods), map comparisons (e.g., SNPs verses microsatellites), gene-gene interaction and gene-environmental interaction studies.
Due to the huge potential benefit to the public health, geneticists and analytical researchers, including statisticians, have focused their efforts on finding genes affecting susceptibility to common, complex disorders such as diabetes, asthma, hypertension, cardiovascular and psychiatric diseases. The transmission of these disorders is complex, the etiologic complexity being increased by the action and interaction of multiple genes and environmental factors. There are other complications such as sporadic cases, incomplete penetrance (i.e., genetically predisposed individuals might not exhibit the disorder) and late age of onset. All these factors increase the difficulty of identifying the genetic components of the trait of interest. Genetic linkage studies are often the first step in finding and cloning a disease gene. Their goal is to locate and, if possible, shorten regions on the genome that are very likely to contain disease susceptibility genes. In many studies, difficulty arises because most genetic data sets are incomplete and investigators want to know how much information in the data is available for the study relative to the amount of information that would have been available if the data were complete. This relative information directly guides the investigator's follow-up strategies (e.g., using more genetic markers with existing DNA samples versus collecting DNA samples from more families), and a misleading measure can lead to a serious waste of human and financial resources as well as a delay in the progress of the underlying genetic studies. The goal of this proposal is to provide reliable measures of such relative information by using the current start-of-the-art statistical techniques.
