Abstract: We introduce a new approach for modeling the influence of a disease process on gene or protein expression patterns. Our emphasis is on the simultaneous, multivariate characterization of expression alterations across large numbers of genes, rather than on the construction of normal/affected differential expression profiles one gene at a time. The key idea is to reconstruct the expression profile for a latent sample of normal control cells corresponding to each disease sample, and then use the displacements between the disease samples and their reconstructed controls to uncover a low-dimensional range of alternative disease progression pathways. The method is easy to implement, and is expected to be widely-applicable to genomic and proteomic studies using a broad range of large-scale assay technologies. We demonstrate the method by applying it to gene expression studies of colon cancer and breast cancer.
Key words and phrases: Cancer, differential expression, dimension reduction, gene expression, tumor progression.