[Symposium 57] Perception and stigma, trans-ethnic genetics, virtual reality treatment of anxiety disorder, and Hyperemesis Gravidarum

[SY-57-02] Transethnic Genetic Etiology of Panic Disorder: Approaches Using Polygenic Scores and Their Machine Learning-Based Classification

*Kazutaka Ohi¹, Takeshi Otowa², Hisanobu Kaiya³, Tsukasa Sasaki⁴, Hisashi Tanii⁵, Toshiki Shioiri¹ (1.Department of Psychiatry, Gifu University Graduate School of Medicine(Japan), 2.Department of Psychiatry, Teikyo University(Japan), 3.Panic Disorder Research Center, Warakukai Medical Corporation(Japan), 4.Department of Physical and Health Education, Graduate School of Education, The University of Tokyo(Japan), 5.Center for Physical and Mental Health, Mie University(Japan))

Keywords:

Panic disorder,Polygenic score,Machine Learning

Panic disorder (PD), one of the core anxiety disorders, is modestly heritable worldwide despite cultural differences across countries. The genetic basis of anxiety disorders overlaps with that of other psychiatric disorders, such as major depressive disorder (MDD), as well as with intermediate phenotypes such as neuroticism, particularly in individuals of European ancestry. First, we have comprehensively investigated the transethnic polygenetic features shared between European individuals with psychiatric disorders and their intermediate phenotypes and Japanese individuals with PD [718 PD and 1,717 healthy controls(HCs)] using several polygenic scores (PGSs) derived from large-scale genome-wide association studies. Second, we have examined whether individuals with PD could be reliably diagnosed by utilizing combinations of multiple PGSs-up to 48- for psychiatric disorders and their intermediate phenotypes, compared with single PGS approaches, using specific machine learning classifiers: logistic regression, neural networks, quadratic discriminant analysis, random forests, and support vector machines. Our results demonstrated that PGSs derived from European studies of anxiety disorders and MDD were associated with PD in the Japanese populations. Among intermediate phenotypes, PGSs for loneliness, neuroticism, and lower cognitive function were also associated with Japanese PD individuals. All five classifiers performed relatively well in distinguishing PD individuals from HCs, with classification accuracy improving as the number of PGSs increased. The greatest areas under the curve at the best PGS combination significantly differed among the five classifiers. Notably, random forests exhibited the lowest accuracy, while support vector machines had higher accuracy than neural networks in classification performance. Our findings suggest that PD shares transethnic genetic etiologies with other psychiatric disorders and related intermediate phenotypes. Moreover, increasing the number of PGS, up to approximately 10, effectively improved the classification accuracy. Among the classifiers tested, support vector machines exhibited the highest accuracy. However, the overall classification accuracy of PD based solely on PGS combinations remained modest.