Multivariate methods aid in pinpointing promising tumor marker candidates from colorectal biopsies

DOI: 10.5584/jiomics.v2i1.79

Abstract

The application of proteomic techniques to the search for disease markers is widely reported nowadays. However, the data rendered by these methods is highly complex and requires mining through statistical methods. Since univariate tests are prone to false positives and require post-test correction, multivariate methods seem more suitable for the task. Here we show an example of their utility, applying both principal component analysis (PCA) and linear discriminant analysis (LDA) to the hydrophobic subproteome of the colorectal mucosa. In order to find proteins specifically altered by colorectal cancer, we compared both the tumor and the adjacent healthy mucosa. PCA followed by variable selection, and corroboration by LDA, pointed out the proteins vimentin and prohibitin as promising candidates for the diagnosis of colorectal tumors.