An information-theoretic approach to high-dimensional pharmacometrics: Applications for interaction analyses in drug development
MetadataShow full item record
The following dissertation develops an information-theoretic computational framework, and shows applications for analysis of high-dimensional datasets such as those routinely encountered when conducting pharmacogenetic/genomic clinical trials. The successful application of information-theoretic concepts in pharmaceutical datasets provides a novel set of pharmacometric tools that may be leveraged to increase learning on genome scale datasets. A series of novel algorithms coded in the javaTM programming language with computational roots in information theory were extended and utilized as the basis for the methodology. Simulations and actual clinical datasets representing a broad range of complexity were utilized to highlight the capabilities of the computational approach on high-dimensional datasets. Additionally, the work was compared to existing methodologies on both theoretical and practical levels. The results suggest that an information-theoretic analytical platform offers an appropriately flexible and computationally efficient basis for performing interaction analyses on high-dimensional data sets. In most cases, the proposed methods performed comparably to, or better than, existing methodologies when the size of the data set was not prohibitive for traditional approaches: These results held across all levels of complexity, and for clinical data as well simulated data. In situations where multiple types of relationships may exist and there is no specific need for a structural parameterization, the information-theoretic approach proposed here may serve as an appropriate analytical platform, capable of detecting novel interactions and informative relationships. The algorithms that have been extended here are computationally efficient enough to allow detection of higher-order relationships in genome-scale data sets on common laboratory computers, negating the need for access to sophisticated computational facilities. Thus, the advancements realized by this work are as much theoretical as they are practical.