Remote computational phenotyping of behavioral and affective dynamics in major depression

点击翻译按钮获取中文摘要

PROJECT SUMMARY / ABSTRACT Major depression is a highly debilitating disorder affecting over 300 million people worldwide. Treatment assignment can involve a lengthy trial-and-error process complicated by symptom heterogeneity. The Research Domain Criteria (RDoC) matrix provides a framework for investigating psychiatric disorders that integrates across multiple levels of analysis. Depressive symptoms are closely linked to the RDoC Positive Valence Systems (PVS) domain, but it is unknown how PVS constructs relate to common depressive symptoms including low mood, anhedonia, and apathy. Computational probes of behavioral and affective dynamics show great promise as a means of ‘computationally phenotyping’ individuals and providing a way to validate PVS constructs in relation to symptom heterogeneity. The ubiquity of smartphones makes them an ideal platform for remote testing. We propose to collect longitudinal data using smartphones for three ‘gamified’ tasks that measure risky decision making, probabilistic reinforcement learning, and reward-effort trade-offs and concurrent fluctuations in affective state. We will establish the reliability of remotely collected computational assays of behavioral and affective dynamics for understanding heterogeneity in depressive symptoms. We will first test a community sample (n=200) both in the lab and remotely by smartphone to verify that behavioral and affective computational parameters have the same relationship to depressive symptoms (low mood, anhedonia, and apathy) in both environments (Aim 1). We will then recruit a large sample of patients with moderate depressive symptoms (n=400) and test them remotely using smartphones for up to 12 months (Aim 2). We will test whether behavioral and affective computational parameters are related to changes in depressive symptoms over time. We will also use data-driven recurrent neural network approaches to identify additional features of our data related to depressive symptoms. Finally, we will collect MRI scans and in-lab data in a subsample of patients (n=200) from Aim 2 and ask whether reward sensitivity and reward prediction error, features of all three tasks, map onto consistent neural circuitry and depressive symptoms (Aim 3). We will test for a mapping between depression subtypes defined by brain network connectivity, behavioral and affective computational parameters, and depressive symptoms. Using computational models, we can bridge between levels of circuits, behavior, and self-report, and test for a mapping onto heterogeneity in symptoms, enhancing our understanding of RDoC constructs and paving the way for more effective and timely interventions to treat depression.