Evaluating the performance of polygenic risk score and machine learning-based classification for the prediction of posttraumatic stress disorder in a south african population

Posttraumatic stress disorder (PTSD) is a complex psychiatric disorder characterized by symptoms of intrusive thoughts, avoidance behaviors, hyper-arousal and negative alterations in cognition and mood. PTSD is unique among psychiatric disorders in that its identification is conditional upon exposure to a traumatic incident. While 50-85% of individuals are expected to encounter at least one traumatic event in their lifetime, the prevailing prevalence of PTSD is estimated to lie between 3.9 and 5.6% globally. This discrepancy serves to highlight the existence of factors granting individuals contingent resilience or vulnerability to developing PTSD. While the biological underpinnings elemental to PTSD remain largely unknown, prior heritability estimates have suggested that the disorder presents a genetic component that interacts with non-genetic factors to confer risk of or resilience to PTSD. This study aims to elucidate the molecular mechanisms underlying PTSD by comparing the predictive performance of a series of PTSD-risk proxies in a uniquely admixed South African population. Polygenic risk score- and machine learning-based predictive approaches will be used to construct, optimize, and subsequently validate models tailored to assess genetic risk in our sample. In addition, we will explore the utility of genetically supported predictor variables by utilizing transcriptome analysis to test the generated risk proxies against potential molecular contributors. Data will be presented on preliminary polygenic risk score models attempting to identify the optimal method to predict PTSD status in our sample population. These findings will add to the growing knowledge base on polygenic risk score- and machine learning-based methods in psychiatric studies, supplement our current research on the genetic mechanisms underlying PTSD, as well as help improve existing analytical capabilities associated with interrogating complex datasets in resource-limited environments.