Fig. 3

Overview of the potential research pathway leading from data to the identification of functional endotypes. First, clinical and biological data have to be collected in the framework of observational cohorts or randomized controlled trials. Critical relevance lies in the collection of samples that allow the implementation of high-throughput biological analyses in a second step. Optimally data from multiple databases are bundled in order to allow a robust subphenotype discovery. In a third step, data are fed into an unsupervised machine learning pipeline, which hopefully identifies clusters of patients in the given multi-dimensional variable constellation. These clusters or subphenotypes have then to be validated in an external prospective cohort, and optimally, a parsimonious model is then elaborated that allows identification of subphenotypes at the bedside with a minimal number of variables. Finally, and as the ultimate goal of phenotyping, a biological correlate or ideally, a treatable trait, is identified for each subphenotype, which can be targeted by means of a specific medication, leading to the transition from a subphenotype to a functional endotype