An Automatic Relevance Determination Method for Identifying the Signature of Stage I Ovarian Cancer in the Mass-Spectrum of Serum Proteins

摘要

Ovarian cancer (OC) occurs in 1 in ～2,500 women in the general population and presents in a late clinical stage (Stage Ⅲ) in more than 80% of patients. This 80% has a 5-year survival rate of only ～35% . If OC could be detected near onset (Stage Ⅰ), in contrast, it is generally believed that the 5-year survival rate could be increased to greater than 90% under existing treatment protocols. Petricoin and co-workers have reported a dramatic breakthrough in detection of Stage Ⅰ OC by applying a proprietary genetic algorithm to mass spectra of serum proteins, correctly classifying ～90% of the "truth " data. Here I report that a Bayesian neural net with automatic relevance determination (ARD) correctly classifies 92% that same data.