Previously, adaptive learning algorithms have been used with immunosignaturing in order to identify disease states in patients. However, in these algorithms the presence of a single disease state is assumed, although in a clinical setting this may not be the case. We propose a novel algorithm based on latent feature identification using beta process factor analysis, in which the binary feature sharing matrix is modified and key comparisons are applied to identify multiple possible underlying disease states. The algorithm is verified using combinations of actual patient immunosignaturing data. The proposed method has a variety of applications, including multi-disease state diagnosis in the clinical setting, multi-biothreat detection in the field, and separation of co-contaminated biological samples.