Unsupervised Data Mining Applications on High Dimensional Gene Expression Time Series in Toxicogenomics

Toxicogenomics, the study of adverse effects caused by toxicants to human health and environment via high-throughput genomics technologies, present promising alternatives to expensive and lengthy animal-based approaches in toxicity testing and risk assessment. Advances in toxicogenomics techniques now enable monitoring cellular activities continuously, under a large range of experimental and biological conditions and providing comprehensive and high-resolution information at molecular levels. The research interests in toxicogenomics center on key issues involving the quantification of molecular toxic effects, the linkage between molecular endpoints and phenotypic ones, the discernment of dose-response and pharmacokinetics relationships, as well as the integration of bioinformatics into predictive toxicology. In particular, the increasingly complex and voluminous toxicogenomics data pose great analytical challenges. The existing bioinformatics tools are incompatible with the high dimensionality and temporal dynamics of the data, possibly leading to unreliable and misinterpretation of the potential toxicity connotation. The objectives of this dissertation are to develop and demonstrate new or improved methodology that better address the challenges and limitations in high dimensional time series toxicogenomics data analysis for critical bioinformatics application such as toxicity mechanism identification, toxicants classification, and for predictive toxicology knowledge discovery. In this study, we develop new or improve bioinformatics data analysis algorithms so that they are capable of processing high dimensional time series toxicogenomics data, therefore better capture and reflect the dynamics of cellular response to toxicants. We also prove the potential and validity of the incorporation of various molecular disturbance/effect quantifiers into various functional toxicogenomics bioinformatics to provide quantitative insights into the toxicant-induced cellular molecular responses at individual gene, specific pathway and system levels. In addition, we demonstrate the effectives of unsupervised bioinformatics tools for mining new, more in depth, much-detailed and fundamental knowledge and understanding of toxicological information at molecular level. This research could generate new information to fill in the urgent knowledge gaps in toxicogenomics that present barriers to the realization of predictive toxicology and make contributions to several fields including toxicology, bioinformatics and environmental science.