The identification of genes responsible for causing cancers from gene expression data has had varied success. Often the genes identified depend on the methods used for detecting expression patterns, or on the ways that the data had been normalized and filtered. The use of gene set enrichment analysis is one way to introduce biological information in order to improve the detection of differentially expressed genes and pathways. In this paper we show that the use of network models while still subject to the problems of normalization is a more robust method for detecting pathways that are differentially overrepresented in lung cancer data. Such differences may provide opportunities for novel therapeutics. In addition, we present evidence that non-small cell lung carcinoma is not a series of homogeneous diseases; rather that there is a heterogeny within the genotype which defies phenotype classification. This diversity helps to explain the lack of progress in developing therapies against non-small cell carcinoma and suggests that drug development may consider multiple pathways as treatment targets.