Abstract #5496: siRNA target selection using machine learning techniques: Genes implicated in aggressive prostate cancer

Abstract

Distinguishing between indolent and aggressive forms of prostate cancer remains a significant clinical problem, existing markers which define these cancers are inadequate. We urgently need to replace these with more reliable and robust alternative biomarkers to spare those with indolent cancers the morbidity of treatment and ensure those with clinically relevant cancers are treated. We have previously used an unsupervised learning technique called latent process decomposition to find additional structure in breast cancer expression microarray datasets. We have now applied this analytical approach to published prostate cancer expression microarray datasets leading to the discovery that prostate cancer may be divided into three prognostically distinct subgroups. Genes overexpressed in the aggressive subgroup were examined for biological function in prostate cancer cell lines using siRNA technology. Using this method we identified two genes, PSM\#946;5 (which encodes a \#946; subunit of the 20s and 26s proteasomes) and C7orf28A (which encodes a novel lysosomal protein) that were of particular interest. These genes, when silenced, significantly reduced growth and viability of prostate cancer cell lines PC3 and 22Rv1 at least in part by causing an increase in apoptosis. Quantitative RT-PCR studies confirmed that PSM\#946;5 is expressed at abnormally high levels in prostate cancer and antibodies raised against C7orf28A show reduced expression in normal prostate. This study illustrates the practical use of unsupervised learning methods in selecting genes of potential prognostic, therapeutic and functional interest.

Citation Information: In: Proc Am Assoc Cancer Res; 2009 Apr 18-22; Denver, CO. Philadelphia (PA): AACR; 2009. Abstract nr 5496.