Biomarker based classification and clustering of Veterans with PTSD

Posttraumatic Stress Disorder (PTSD) is both impactful on a societal and individual basis. Better treatments are needed for our Veterans, and methodological advancements can facilitate progress in the understanding and treatment of PTSD. With the dramatically reduced cost of collection and analysis of DNA there is now many large repositories that contain relatively comprehensive genetic data. Furthermore, on the individual level, many Veterans have genetic data available through commercial (e.g., 23 and Me) or research (e.g., the Million Veteran Program) entities. It is possible that access to their own genetic data will allow people to leverage the latest research developments in the aims of receiving state-of-the-art personalized medicine. New algorithmic advances can allow this data to be utilized for improving our understanding of conditions, such as PTSD. Machine learning provides a way to better identify those with PTSD through methods such as boosted trees or deep learning models. Clustering techniques can provide a way to clarify homogenous subgroups within Veteran samples. This project looks to take a supervised (aka, classification) and unsupervised (aka, clustering) analytical approach to better understand PTSD using genetic and neuroimaging data from publicly available data repositories. For this work, features are selected based on the latest advancements in the literature and models are selected through rigorous empirical evaluation. Treatment for PTSD, and other disorders, is potentially limited by classifications that are based on symptom reports such as used in DSM. This traditional approach creates biologically heterogeneous samples. We propose applying advanced analytical tools to empirically derived biomarkers to create homogeneous biologically-based groupings.