Bioinformatics for post-traumatic stress

Project Summary/Abstract Maladaptive complications following trauma, including post-traumatic stress (PTS), are highly prevalent in both veterans and civilians, and have been difficult to accurately diagnose, manage and treat. Debate regarding diagnostic criteria and the need to represent the full spectrum of inter-connected features contributing to psychopathology has spawned the development of the Research Domain Criteria (RDoC) by the National Institute of Mental Health (NIMH). RDoC is a developing framework to help guide the discovery and validation of new dimensions of mental health disorders and their relationships to underlying biological mechanisms. NIMH now has a rich federated database that currently houses raw data from RDoC-sponsored clinical research, and clinical trial data from the National Database of Clinical Trials (NDCT) with information that may help to unlock the complex and overlapping relationships between symptoms of PTS and the underlying biomarkers to fuel improvements on diagnostic and therapeutic frameworks for trauma recovery. The proposed project will apply bioinformatics and machine learning analytical tools to these large, heterogeneous datasets to identify and validate new research dimensions of trauma-related psychopathology and treatment response trajectories and their predictors. Aim 1 will develop an in silico trauma patient population by integrating data from diverse sources, including cross-sectional and observational longitudinal clinical studies housed within available data repositories for trauma and other related mental health research. Data will include medical history, demographics, diagnostic tests, clinical outcomes, psychological assessments, genomics, imaging, and other relevant study and meta-data. Aim 2 will identify multiple dimensions of PTS diagnostic criteria, using a combination of unsupervised dimension-reduction statistical methods, internal and external cross-validation, and supervised hypothesis testing of predictive models to understand the heterogeneous subtypes of PTS. Aim 3 will deploy unsupervised machine learning methods, such as topological data analysis and hierarchical clustering, to identify unique clusters of patients based on symptomatology to develop clustering methods for precision mapping of PTS patients based on disease severity. Aim 4 will use supervised machine learning techniques for targeted predictive analytics focused on identifying treatment responders from the NDCT, and identification of latent variables that predict treatment response. The results of the proposed research project will greatly enrich the field of computational psychiatry research to identify conserved dimensions associated with the complex relationships of psychopathology and precision treatment planning following exposure to traumatic events.