Sample A: Cover Page of Thesis, Project, or Dissertation Proposal

statistical criterion, but it appears to be unable to classify the adenocarcinomas because the
Euclidean Sum of Squares is two fold larger than for the (smaller N) squamous and normal
clusters. The ProbeSets KRT17, EMP2 and MFAP4 which were selected by both criteria,
demonstrate that one method by which to avoid such false positives: that is, select an average
probe gain threshold criteria of ProbeSets whose aggregate gain improves on the average probe
gain. Thirteen of 29 ProbeSets demonstrating an average probe gain had an aggregate gain ratio
that improved on the average probe gain ratio. In contrast to the average probe gain list of 29
genes, half of this list (6 of 13) demonstrates enhanced ProbeSet information gain; having a
ProbeSet gain ratio greater than 0.8. The 7 additional genes for demonstrate a close relationship
to the underlying biology.
Such effective down selection, at the ProbeSet level, may be a reasonable endpoint if the goal is
to produce an effective and simple diagnostic test. However, this is a limited approach if the goal
is to understand the biological mechanisms of disease. Based upon the classification
performances, presented in Figure 5.1, 18 Genes were identified were selected as a candidate list
(Table 5.1) to explore the biological function with respect to NSCLC.
Literature Validation of Biological Process
We observed significant down regulation of TACC1 in both forms of NSCLC, which is supported
by the literature [15-17]. There are 3 TACC human genes, which appear to be important for
cellular division and organization [18], with orthologues reported in Mus. musculus, Drosophila
melangaster, and Xenopus laevis [18]. Proper localization of TACC during cytokinesis appears
to be dependent upon phosphorylation by aurora kinases [15, 19] and may posses a critical
function in cell cycle control [17]. Appropriate cell division is a prerequisite in the propagation
125