Background: The low specificity of the prostate-specific antigen (PSA) test makes it a poor biomarker for early detection of prostate cancer (PCA). Because single biomarkers most likely will not be found that are expressed by all genetic forms of PCA, we evaluated and developed a proteomic approach for the simultaneous detection and analysis of multiple proteins for the differentiation of PCA from noncancer patients.
Methods: Serum samples from 386 men [197 with PCA, 92 with benign prostatic hyperplasia (BPH), and 96 healthy individuals], randomly divided into training (n = 326) and test (n = 60) sets, were analyzed by surface-enhanced laser desorption/ionization (SELDI) mass spectrometry. The 124 peaks detected by computer analyses were analyzed in the training set by a boosting tree algorithm to develop a classifier for separating PCA from the noncancer groups. The classifier was then challenged with the test set (30 PCA samples, 15 BPH samples, 15 samples from healthy men) to determine the validity and accuracy of the classification system.
Results: Two classifiers were developed. The AdaBoost classifier completely separated the PCA from the noncancer samples, achieving 100% sensitivity and specificity. The second classifier, the Boosted Decision Stump Feature Selection classifier, was easier to interpret and used only 21 (compared with 74) peaks and a combination of 21 (vs 500) base classifiers to achieve a sensitivity and specificity of 97% for the test set.
Conclusions: The high sensitivity and specificity achieved in this study provides support of the potential for SELDI, coupled with a bioinformatics learning algorithm, to improve the early detection/diagnosis of PCA.