We are collaborating with two independent university investigators who are generating clinical and genomic data from cancerous and normal prostate tissues. This research effort is designed to develop RFE-SVM tools to delineate prognosis, disease progression and treatment response of prostate cancer patients based on clinical information datasets as well as DNA microarray datasets.

A Clinical Dataset with 158 patients (79 Transition Zone patients and 79 Peripheral Zone patients) was analyzed, using accepted variables that influence patient outcome, to determine if the SVM techniques could be used as an effective clinical outcome prognosis tool.

A microarray experiment was performed on radical prostatectomy tissues resulting in 7,129 gene expression measurements for each sample across a population of 67 tissues removed from 26 patients. The objectives of the study were to find a genetic link between where the cancer was located in the prostate, in what type of tissue the cancer was located, and the severity and the recurrence potential of the disease. The initial experiments were limited to discerning the genetic differences between 42 tissues of Benign Prostate Hyperplasia (BPH) and the highly malignant grade four cancers. Ten genes were identified that had greatest correlation with the two datasets. However, a subset of only two genes predicted the separation of BPH and grade four cancer patients perfectly, using the leave-one-out method. BIOwulf holds the exclusive rights to a patent application on this discovery. This study will be expanded to include other regions and grades of prostate cancer. The ability of our Machine Learning technologies to discriminate signal below the threshold for many conventional methods allowed use of 60% more microarrays than conventional methods. The lower susceptibility to spurious noise in BIOwulf¹s analysis enabled the use of experiments that even Affymetrix had labeled as unusable. The higher level of sensitivity in this technology allows interpretation of much of the lower signal data and greatly improved results.