ABSTRACT Suicidal thoughts and behaviors (STBs) are growing more prevalent among adolescents; despite these alarming trends, researchers have been hampered in their efforts to identify antecedents of STBs because of the transient nature of suicidal impulses that are unlikely to be captured during a clinical or laboratory assessment. Furthermore, research in this area has focused primarily on time-invariant factors (e.g., gender) and self-disclosed information, which greatly limits our understanding of the neurobiological and psychosocial mechanisms underlying the etiology and maintenance of STBs. Advances in real-time monitoring technology, including mobile apps, provide an unprecedented opportunity to continuously measure key behaviors relevant for understanding suicide risk (e.g., social interactions, sleep) outside of the laboratory for the purposes generating digital phenotypes of STBs. Moreover, statistical approaches such as machine learning are ideal for handling high-dimensional data across different constructs (e.g., clinical, digital, neurobiological) and are increasingly being used in the context of improving prediction of STBs. Thus, the overarching goal of this supplement is to collect and integrate digital phenotypes with neurobiological phenotypes in a machine learning framework to identify multi-level factors associated with the etiology and maintenance of STBs in a high-risk sample: depressed adolescents. Specifically, we will build on the existing infrastructure of the parent grant—which focuses on characterizing the stress-related neurobiological trajectories using a multi-level approach in a sample of depressed adolescents—by seeking to identify multi-level predictors and trajectories of STBs in this high-risk sample and to compute deviations from normative phenotypes and trajectories computed from a low-risk comparison group. We will use machine learning algorithms to identify the constellation of factors that best predict likelihood of engaging in STBs by Time 3 among the depressed adolescents (Aim 1); we will also identify the factors that best predict trajectories of STBs based on changes from Time to Time 3 among the depressed adolescents (Aim 2); we will also test whether deviations from normative phenotypes and trajectories (computed from data in the healthy controls) are better predictors of STBs (Aim 3). In accordance with NOT-MH-19-026 (“Administrative Supplements for NIMH Grants to Expand Suicide Research”), this approach addresses current barriers in our understanding of the mechanisms of action underlying suicide risk by collecting ecologically valid measurements of suicide-relevant behaviors and by fostering advanced statistical methods for multi-level and cross-construct integration.