Gabriel: Platform for explicit, systematic analysis of high-throughput genetic data

High-throughput genetic tools such as cDNA Microarrays have been used extensively to investigate biological and clinical questions. Genome-wide gene expression profiles and signatures are produced to characterize biological and clinical states. Many microarray analytical approaches have been developed to organize data. However, the basis for threshold choice decisions was often non-transparent. Additional information was often applied after the analysis, subject to a user's non-transparent judgment. In addition, commonly only one threshold choice was applied, and threshold choices were not systematically examined. To address these issues, in my thesis, I asked three key questions: (1) What should be the features for developing a generic computational analysis paradigm for high throughput genetic analysis that facilitates explicit and systematic threshold choices? (2) How does threshold choice influence biological conclusions? (3) Does a conclusion with a wider range of thresholds with statistical significance give us more confidence in its biological validity?; To study these questions, I designed a GABRIEL (Genetic Analysis By Rules Incorporating Expert Logic) approach, and we developed the GABRIEL platform that provides the following features: (1) Threshold choices are made explicitly; (2) Additional Information incorporated as a GABRIEL pattern and applied before the analysis; (3) GABRIEL facilitates systematic application of a range of threshold choices. To test the general applicability of the GABRIEL approach, GABRIEL was applied to study several biologically important questions, including pathways of antibiotic biosynthesis in Streptomyces, molecular fingerprints characteristic of replicative senescence in mammalian cells, and the molecular signatures in hepatocellular carcinoma. Novel biological findings have been made by using GABRIEL.; To study the second question, I evaluated biological conclusions under different threshold choices and found that threshold choices can influence the biological conclusions from microarray analysis. The results suggest the importance of making threshold choices explicit and systematic.; To study the third question, I proposed a new concept, robustness index, to assess the robustness of a biological conclusion across threshold choices. By using independent confirmation data, I show that robustness index provides a measure of the confidence we should have regarding how well a biological conclusion approximates the biological reality.