This competing renewal application seeks to conclude a consecutive series of four separately funded studies, which enabled the gathering of 15+ years of data, on average, on young, depressed patients from childhood to young adulthood. Our longitudinal developmental data, reflecting multiple domains of functioning, can yield actionable information about which risk and protective variables/domains best predict clinical and functional outcomes of juvenile-onset depression (JOD), which is a particularly severe depression phenotype. Our AIM 1 is to deposit in the National Data Archives (NDA) the data from the first three studies, which will complement the mandated archiving of data from the most recent (fourth) project. Thereby, these unique data will be accessible to future analyses by other researchers. Because commonly used modelling approaches cannot accommodate our questions and the complexity and size of our data base, our AIM 2 is to demonstrate the novel application of two approaches from the machine learning toolbox (probabilistic graphical models and ensemble learning methods) to predict JOD outcomes. To enable researchers to fully utilize the data that will be deposited in the NDA, we will release the Python code packages we develop for AIM 2 as well as the code for downloading and properly organizing the related information. The first study, a Program Project started in 2000, included 7- to 14-year-old young patients (probands; n=711) from 23 mental health facilities across Hungary, whom we diagnosed as having a DSM-IV depressive disorder; biological siblings of probands (n=301) were also recruited. Portions of the samples were later enrolled in three consecutive studies, which also included never depressed controls. The most recent project, ended in 2021 when participants were in their mid-20’s to early 30’s, included 308 probands, 229 siblings of probands, and 160 controls. (The reduced sample sizes, compared to prior ones, were due to funding limits). Across the four projects, close to 1,100 individuals had two or more assessments covering a large array of domains and variables: key constructs were assessed repeatedly and in multiple ways. To implement AIM 1, our longitudinal data will be harmonized with NDA structures and definitions and then deposited. To implement AIM 2, developmentally-framed hypotheses will guide the novel application of machine-learning approaches to JOD outcomes under two scenarios: for outcomes with a variety of well-known predictors (e.g., recurrent depression) but scant information about the interrelationships among them and about which are “genuine” predictors, we will implement probabilistic graphical models; for outcomes the predictors of which are not well established and/or are supported by equivocal information (e.g., emotion regulation competence in daily life), we will use ensemble learning methods. The new knowledge we will generate about JOD will have conceptual implications, will inform efforts to p...