Identifying subgroups from a severely heterogeneous population is major challenge for Big Data. Different clustering methods optimize differently and consequently capture different aspects of relatedness in the population. Since there is not a one size fits all solution, and no gold standard, the selection of a clustering method can be daunting and problematic. Our interdisciplinary team is working towards the development of interactive ensemble methods for clustering Big Data.In this first year, we have begun to lay the methodological foundation through the development of a non-parametric bootstrapping approach to estimate the stability of a clustering method. We have developed two novel approaches to bootstrapping stability, and accompanying visualizations, that accommodate different model assumptions, which can be motivated by an investigator’s trust (or lack thereof) in the original data. Our approaches outperform state of the art methods for simulation and real data sets of moderate size. A long term vision of our work is to extend this bootstrapping approach to improve classification and diagnosis of mood disorders, in particular bipolar disorder and major depressive disorder, using data from the UK Biobank. This endeavor would require automated feature selection, sophisticated visualizations, and methods that accommodate mixed data, while retaining valuable clinical interpretations. This project is motivated by the hypothesis that a more precise and personalized classification of mental health disease can be obtained through the development of novel clustering methods that identify clinically significant structures with large population data sets.