Enhancing the discriminatory power of polygenic scores for ADHD and autism in clinical and non-clinical samples.

Abstract

Background: Polygenic scores (PGS) are widely used in psychiatric genetic associations studies due to their predictive power for focal outcomes. However, they lack discriminatory power, in part due to the high degree of genetic overlap between psychiatric disorders. The lack of prediction specificity limits the clinical utility of psychiatric PGS, particularly for diagnostic applications. The goal of the study was to enhance the discriminatory power of psychiatric PGS for two highly comorbid and genetically correlated neurodevelopmental disorders in ADHD and autism spectrum disorder (ASD).

Methods: Genomic structural equation modeling (GenomicSEM) was used to generate novel PGS for ADHD and ASD by accounting for the genetic overlap between these disorders (and eight others) to achieve greater discriminatory power in non-focal outcome predictions. PGS associations were tested in two large independent samples - the Philadelphia Neurodevelopmental Cohort (N = 4,789) and the Simons Foundation Powering Autism Research for Knowledge (SPARK) ASD and sibling controls (N = 5,045) cohort.

Results: PGS from GenomicSEM achieved superior discriminatory power in terms of showing significantly attenuated associations with non-focal outcomes relative to traditionally computed PGS for these disorders. Additionally, genetic correlations between GenomicSEM PGS for ASD and ADHD were significantly attenuated in cross-trait associations with other psychiatric disorders and outcomes.

Conclusions: Psychiatric PGS associations are likely inflated by the high degree of genetic overlap between the psychiatric disorders. Methods such as GenomicSEM can be used to refine PGS signals to be more disorder-specific, thereby enhancing their discriminatory power for future diagnostic applications.