A significant challenge facing computational cell population discovery tools for flow and mass cytometry is the cluster matching and annotation problem. The cluster matching problem is the challenge of identifying and labeling biologically identical clusters of cells discovered across independent samples. The annotation problem is the challenge of labeling cell populations with their phenotypes. Most existing methods tackle the former by standardizing and multiplexing data across samples prior to clustering, thereby ensuring each analyzed sample has the same number of clusters and the same cluster labels. This approach often fails in the presence of biological and technical variability and with larger data sets where methods come up against computational limitations. The latter problem usually requires significant user intervention in order to annotate discovered cell populations with their phenotypes. Here we present a new methodology entitled FAUST (Full Annotation Using Shape Constrained Trees) that tackles these problems by working in phenotypic space; standardizing and partitioning the space across samples and using the discovered phenotypes to match discovered cell populations across independent samples. Here we demonstrate how FAUST can be used for biomarker discovery across disparate data sets. We apply FAUST to perform cell population discovery across three cancer immunotherapy trials. We show that FAUST recovers expected biology by identifying PD1 expressing T cells as a correlate of outcome, but also that it makes new discoveries, by independently identifying cell populations with monocyte & dendritic cell phenotypes across the trials that predict response to therapy at baseline.