Identification and Validation of Brain Relevant Blood-Based Biomarkers for Suicide in Major Depressive Disorder

Suicide is a complex mental and public health problem. Despite the progress in our understanding of the neurobiology of suicide, we currently lack precise clinical or molecular tools to help us predict suicide risk and reduce the high suicide rates across the country. Major depressive disorder (M.D.D) is the most common diagnosis present in individuals who die by suicide. However, not all M.D.D patients exhibit suicidal behaviors or are otherwise at-risk of suicide. In addition to M.D.D, genetic and environmental factors can also increase the predisposition to develop suicidal behaviors and ultimately to death by suicide. In the context of M.D.D there are currently no validated biomarkers to clinically assess short term suicide risk. The main goal of our proposal is to develop suicide specific molecular biomarkers in blood using a combination of a) post-mortem blood samples to study suicide-specific biomarkers and b) clinical blood samples from living M.D.D patients assessed for suicidal behaviors at baseline and follow-up times. Blood-based transcriptomic and epigenetic data will be used along with clinical data for both cohorts. For the post-mortem cohort, death by suicide will be the main outcome investigated. For the clinical cohort suicidal behaviors will be investigated. Clinical data for the living cohorts will also include questionnaires such as HAM.D., ACE and C-SSRS, as well as demographic data. Blood will be used to study molecular biomarkers in the form of transcriptomics (gene expression), and epigenomics (methylation). Multimodal data will be integrated using machine learning prediction algorithms to identify molecular biomarkers for suicide (post-mortem data) and severe suicidal behaviors and suicide in the clinical cohorts. For prediction, single modality models (methylation and gene expression) will be trained and evaluated using the leave one out cross-validation (LOOCV) method and predictive results of all the LOOCV iterations will be aggregated together to get an overall AUC (area under the curve) value. After feature selection for each modality (methylation and gene expression), we will build the multimodal machine learning model-based integration strategy, this approach could preserve unique information from different modalities and provide complementary information across different types of features. This novel multimodal approach should provide a more comprehensive understanding of both immediate and dynamic (gene expression) and enduring molecular changes (methylation) associated with suicide risk therefore enhancing the identification and characterization of predictive biomarkers. The proposed research project will also establish a collaborative effort integrating clinical professionals, molecular biologists and computational biologist in a team aiming to identify and validate suicide specific blood biomarkers for M.D.D patients. In summary, this project will allow us to determine how multimodal suicide-specific biomarkers, identified in post-mortem blood could enhance suicide risk assessment predictors in clinical at-risk populations.