Abstract
Psychiatric polygenic score (PGS) associations are likely inflated by the high degree of genetic overlap between psychiatric disorders. Methods such as Genomic Structural Equation Modeling (GenomicSEM) can be used to refine PGS signals to be more disorder-specific, thereby enhancing their discriminatory power for future diagnostic applications.
Introduction
In psychiatric genetics, Polygenic Scores (PGS) serve as a summary of an individual's genetic liability for a specific condition. However, a significant hurdle to their clinical application is a lack of discriminatory power. Due to widespread pleiotropy—where the same genetic variants influence multiple traits—an ADHD score often predicts major depression or anxiety almost as effectively as it predicts ADHD. This study aims to untangle these overlapping signals to create "disorder-specific" scores that can better differentiate between conditions.
Methods: Genomic Structural Equation Modeling
To address the issue of genetic overlap, the researchers employed Genomic Structural Equation Modeling (GenomicSEM). This statistical framework allows for the decomposition of genetic variance into shared and unique components across multiple disorders.
The team utilized GWAS summary statistics for 10 disorders, including ADHD, Autism Spectrum Disorder (ASD), Schizophrenia, and Bipolar Disorder. They identified a hierarchical structure consistent with the Hierarchical Taxonomy of Psychopathology (HiTOP) model, grouping disorders into three primary factors: "Internalizing," "Thought," and "Neurodevelopmental."
By accounting for the variance shared across these factors, the researchers could estimate the "unique" genetic signal for ADHD and ASD. They generated three types of scores for comparison: traditional "unfiltered" PGS, disorder-specific PGS (focusing on unique signals), and a Neurodevelopmental PGS (representing shared risk between ADHD and ASD).
Key Insights from Clinical and Population Samples
The study validated these scores using two distinct datasets: the Philadelphia Neurodevelopmental Cohort (PNC) for the general population and the SPARK cohort for clinical ASD cases.
Reducing Genetic "Noise"
A primary finding was that traditional GWAS hits for ADHD and ASD are largely pleiotropic. By isolating the "unique" signal, the researchers discovered that many genes previously associated specifically with ADHD actually belong to a broader neurodevelopmental risk factor. The GenomicSEM-tuned scores were significantly "cleaner"; for instance, the refined ASD score showed zero significant associations with non-focal outcomes in the PNC sample, whereas the traditional score flagged six unrelated disorders.
Clinical Specificity and Comorbidity
In the SPARK clinical sample, where all participants have an ASD diagnosis, the GenomicSEM-tuned ADHD score successfully predicted which children also had comorbid ADHD. In contrast, the traditional ADHD score was confounded by the underlying ASD risk shared by all participants. Furthermore, the "Neurodevelopmental" factor score proved highly effective at predicting co-occurring status (individuals having both ADHD and ASD), suggesting that shared genetic risk is a distinct and measurable entity.
Limitations and Future Directions
While promising, the study notes several limitations:
- Ancestry Bias: The analysis was restricted to individuals of European ancestry, highlighting a critical need for more diverse GWAS data to ensure equitable clinical utility.
- Developmental Factors: As the participants were primarily children and adolescents, future research must determine if these genetic signals remain stable or change in predictive power into adulthood.
- The Specificity-Power Trade-off: Refining scores for specificity often results in a slight decrease in total predictive power ($r^2$). Finding the optimal balance between power and accuracy remains a key challenge for future diagnostic tools.
Conclusion
By leveraging GenomicSEM to filter out shared genetic noise, this research demonstrates a path toward more precise psychiatric polygenic scores. These refined tools offer improved discriminatory power, moving the field closer to using genetic data for meaningful clinical differentiation and personalized intervention.
Summary updated