We have developed an algorithm for the classification of healthy vs. disease whole serum samples using mass spectrometry. We employ a probabilistic classification algorithm built upon a dimension-reduced linear discriminant analysis. On one prostate cancer and three ovarian cancer SELDI datasets, we achieve sensitivity, specificity, and positive predictive values above 97%. This approach is computationally efficient and, perhaps most exciting, can provide clues as to the molecular identities of differentially-expressed proteins and peptides.

Matlab code (Q5)

(Collaborative work with Ryan Lilien and Bruce Donald)