Bioinformatics analysis of 'omics' data towards cancer diagnosis and prognosis

Despite important advances in cancer research in recent decades, an accurate diagnosis and prognosis of cancer remains a formidable challenge to date. In this dissertation, several bioinformatics analyses have been developed for identifying new diagnostic/prognostic signatures utilizing datasets derived from recent high-throughput screening techniques including DNA and protein microarray. In the first analysis we derived an outcome signature from estrogen signaling pathway to predict breast cancer prognosis. This signature successfully predicted patient outcome in multiple patient cohorts as well as ER+ and tamoxifen-treated sub-cohorts. The second part of my thesis focused on applying genetic programming for cancer classification. This approach can automatically select a handful of discriminative genes from gene expression data and produce comprehensible yet efficient rule-based classifiers. In the third analysis, we developed non-invasive diagnostic tools for prostate cancer diagnosis. Two different signatures were yielded from phage peptide microarray system and q-PCR urinary data, respectively. These signatures have the potential to improve specificity and sensitivity of prostate cancer diagnosis. Last, an integrative model was developed for culling a molecular signature of metastatic progression in prostate cancer from proteomic and transcriptomic data. Differential proteomic alterations between localized and metastatic prostate cancer, which were concordant with transcriptomic data, served as a predictor of clinical outcome in prostate cancer. This signature was also predictive of clinical outcome on other solid tumors, suggesting common molecular machinery in aggressive neoplasms. In summary, these bioinformatics analyses of cancer 'omics' data have led to several important findings that may ameliorate cancer diagnosis and prognosis.