Beta process based adaptive learning for immunosignature microarray feature identification

We propose a latent feature model for immunosignature random peptide microarray data using beta process factor analysis to identify relationships between patients and infectious agents. The method uses Bayesian nonparametric adaptive learning techniques that allow for further classification if additional patient data is received, and new relationships between patients and disease states are obtained. In addition to feature discovery, this methodology can also detect biothreat agents on the fly. Using experimental immunosignature microarray data, we demonstrate the identification and classification of underlying relationships between patients with different disease states.