Shortcuts in genome-scale cancer pharmacology research from multivariate analysis of the National Cancer Institute gene expression database.
Application of a soft multivariate statistical procedure, called PLS, partial least squares modelling in latent variables or projections to latent structures, allows extensive exploitation of the enormous amount of information embedded in the National Cancer Institute gene expression and antitumour screen databases. Interpretation of the statistical results provides new significant biological insights such as classification of human tumour cell lines based on their gene expression patterns, evaluation of the influence of gene transcripts on drug efficacy and assessment of their selectivity for classes of compounds which act by the same mechanism, and identification of uncharacterized gene expression targets involved in cancer chemotherapy. Among them, the transcripts GC11121, GC17689, and GC18564 (unknown gene products extremely selective for RNA/DNA antimetabolites) are indicated by the present work as deserving high priority in future molecular studies.