Integrative Bioinformatics Approaches toward Systems-level Understanding of Breast Cancers

Genome-wide expression profiling technologies, such as microarray expression and next-generation sequencing, have allowed unprecedented opportunities to study complex diseases at systems-level. However, the ever increasing amounts of high-throughput genomic data are extremely heterogeneous and each individual experiment provides a different aspect of the phenotype of interest. Meanwhile, numerous bioinformatics tools have been developed for genomic data analysis. Current methods for integration of data and analysis tools are scattered. Developing a systematic approach for the integration of various experiments would greatly benefit the research community.;In this thesis, we present a variety of integrative data-driven bioinformatics strategies to facilitate the data analysis and derive biological knowledge from gene expression data of breast cancer. First, we show that the use of meta-analysis on multiple microarray profiles of estrogen receptor positive and estrogen receptor negative breast cancers reveals important biological functions not found from the individual analysis. By applying a network analysis, we identify the change of gene expression between Luminal A and Luminal B breast cancer subtypes and genes representing the change. Next, we demonstrate a bioinformatics strategy to detect genes that play important roles in endocrine resistance in estrogen receptor positive breast cancers. By combining the analyses of differentially expressed genes, enriched gene set, co-expressed genes, the expression of drug-treated cancer cell lines, and clinical information, we demonstrate how our proposed strategy identifies the key genes in tamoxifen-resistant tumors and the potential new therapeutics against the resistance. Lastly, by using matched mRNA and microRNA expression we develop an integrative approach for the prediction of important transcription factors and microRNAs that are involved in dysregulated pathways in breast cancer. Our method employs random forests and robust rank aggregation to derive a reliable importance ranking for candidate regulators predicted by other bioinformatics tools. In conclusion, this thesis demonstrates that the proposed integrative bioinformatics strategies can efficiently combine heterogeneous genomic data and provide new insights on breast cancers.