Integrating genomics and EHR to accelerate precision psychiatry

Precision psychiatry aims to provide more targeted prevention, personalized risk prediction, and tailored treatments by stratifying patients based on a variety of risk factors, including genetic biomarkers. However, three major factors limit the translation of genomic findings to clinical practice, namely: (1) genetic predictors are trained largely on Eurocentric genome-wide association studies (GWAS) distinguishing disorder cases from controls, rather than on clinical outcomes with greater relevance to patients and clinicians, (2) genomic predictors are often general indices of polygenic risk capturing average effects across disorder subtypes, limiting their ability to more precisely parse heterogeneity in patient profiles, and (3) translational models often fail to capture complex nonlinear relationships, including interactions among genetic predictors or between genetic and clinical risk factors. The proposed work is designed to address each of these gaps. In this proposal, Dr. Tubbs will derive etiological insights into more clinically-relevant psychiatric phenotypes and build translational models to predict these traits in large diverse samples by integrating genomic data with information contained in electronic health records (EHRs). Specifically, Dr. Tubbs will develop phenotyping algorithms for defining three clinical course traits (age at onset, treatment response, and hospitalization), perform multi-ancestry GWAS meta-analyses, and assess phenotyping quality using genetic validation (Aim 1). Dr. Tubbs will also apply a novel approach for dissecting heterogeneity by identifying cross-disorder latent genetic traits using publicly-available GWAS summary statistics (Aim 2). Finally, he will apply a state-of-the-art SuperLearner framework to predict clinical course traits using multi-dimensional polygenic scores (PGS) and EHR-derived risk factors (Aim 3). Together, these aims will uncover novel insights into trans-diagnostic etiology and build translational models for precision psychiatry, while serving as a foundation for a future R01 grant proposal. With his existing skills in statistical/psychiatric genomics, demonstrated productivity, and strong mentorship team, Dr. Tubbs is well-positioned to complete the proposed research and training objectives. Through a combination of coursework, mentorship, and leadership activities, the proposed training plan will strengthen the candidate’s skillset in four key areas: advanced EHR phenotyping, machine learning fundamentals, clinical-translational research principles, and grant-writing/leadership. Dr. Tubbs will continue to be mentored by Drs. Tian Ge and Jordan Smoller, while receiving critical support towards his research and training goals from several collaborators with expertise in EHR phenotyping (Dr. Lea Davis), statistical genetics (Dr. Cathryn Lewis, Dr. Hailiang Huang), and psychiatric machine learning (Dr. Jyotishman Pathak, Dr. Colin Walsh). Overall, this K01 award will be pivotal in Dr. Tubbs’ progress towards establishing an independent research career focused on precision psychiatry.