Imputed Gene Expression versus Single Nucleotide Polymorphism in Predicting Gray Matter Phenotypes

Jiayu Chen, Zening Fu, Armin Iraji, Vince D. Calhoun, Jingyu LiumedRxiv 2023 · DOI

Abstract

Genetics plays a critical role in psychiatric disorders, making the prediction of psychiatric traits from genetic data a key goal for early detection and intervention. This study explores the utility of genetically-regulated expression (GRE)—imputed gene expression reflecting tissue-specific regulatory effects—compared to traditional single nucleotide polymorphism (SNP)-based polygenic risk scores (sPRS). Using a cohort of 34,149 individuals from the UK Biobank, we assessed the prediction accuracy of gene-based polygenic risk scores (gPRS) across 13 schizophrenia-related gray matter networks. Our results demonstrate that gPRS consistently outperforms sPRS, particularly in studies with smaller training sample sizes, suggesting that GRE is a more effective genetic variable for brain phenotype association and prediction.

Introduction

Predicting psychiatric traits and brain phenotypes from genetics is a "holy grail" for early clinical intervention. Traditionally, researchers use Single Nucleotide Polymorphisms (SNPs)—tiny variations in a single DNA building block—to create Polygenic Risk Scores (PRS). However, SNPs have a major drawback: most individual variations have tiny, almost undetectable effects on complex traits like schizophrenia. To achieve accurate predictions, massive datasets (often over 100,000 participants) are typically required to overcome genetic "noise." This requirement is a significant hurdle for brain imaging studies, where collecting MRI scans is expensive and difficult.

This research investigates whether Genetically-Regulated Expression (GRE) can provide a more efficient alternative. Rather than treating every DNA variation as an independent data point, GRE aggregates the effects of multiple SNPs to predict how much a specific gene will be expressed in a specific tissue, effectively filtering biological noise.

Methods and Data

The researchers compared two primary approaches:

  1. sPRS (SNP-based): The standard method summing the effects of individual DNA variants.
  2. gPRS (Gene-based): A method using imputed gene expression levels calculated via MetaXcan and the GTEx database.
Table 1: Brain tissues and the number of genes for which GRE was computed across 13 available brain regions.

The study leveraged data from 34,149 individuals in the UK Biobank cohort. The target phenotypes were 13 specific gray matter networks previously identified as structurally relevant to schizophrenia. GRE was computed for 56,348 genes across 13 brain tissues, aligning the genetic input more closely with the biological output of brain structure.

Figure 1: Visualization of the 13 schizophrenia-related gray matter networks used as target phenotypes for assessing genetic associations.

Key Results

The study found that gPRS outperformed sPRS across all 13 brain phenotypes tested. Key findings include:

  • The Accuracy Gap: At a training size of ~33,000, gPRS reached a mean accuracy (correlation) of ~0.14, while sPRS lagged at ~0.11.
  • The "Sweet Spot": The improvement was most significant when training samples were below 15,000. In this range, gPRS improved prediction accuracy by a margin of 0.019 to 0.029 over the SNP-based method.
  • Biological Filtering: By focusing only on SNPs known to regulate gene expression (eQTLs), the gPRS model ignores thousands of irrelevant genetic variations that clutter standard SNP-based analysis.
  • Heritability: All 13 brain networks were found to be significantly heritable (h² ranging from 0.28 to 0.35), confirming a strong underlying genetic signal.

Discussion and Future Directions

These findings support the use of GRE as a primary genetic variable in brain phenotype studies, especially when sample sizes are limited. While gPRS shows clear advantages in the current data range, the authors note that SNP-based methods might eventually close the gap in massive datasets (e.g., millions of participants) because they consider the entire genome, including non-regulatory regions.

Future work should explore non-linear models, such as deep neural networks, and confirm if these advantages hold true for other imaging modalities, such as functional MRI, or different psychiatric conditions. For now, gPRS offers a powerful tool for gaining deeper biological insights with smaller, more specialized medical datasets.

Summary updated

Abstract

Genetics plays an important role in psychiatric disorders. A clinically relevant question is whether we can predict psychiatric traits from genetics, which holds promise for early detection and tailored intervention. Imputed gene expression, also known as genetically-regulated expression (GRE), reflects the tissue-specific regulatory effects of multiple single nucleotide polymorphisms (SNPs) on genes. In this work, we explored the utility of GRE for trait association studies and how the GRE-based polygenic risk score (gPRS) compared with SNP-based PRS (sPRS) in predicting psychiatric traits. A total of 13 schizophrenia-related gray matter networks identified in another study served as the target brain phenotypes for assessing genetic associations and prediction accuracies in 34,149 individuals from the UK Biobank cohort. GRE was computed leveraging MetaXcan and GTEx tools for 56,348 genes across 13 available brain tissues. We then estimated the effects of individual SNPs and genes separately on each tested brain phenotype in the training set. The effect sizes were then used to compute gPRS and sPRS in the testing set, whose correlations with the brain phenotypes were used to assess the prediction accuracy. The results showed that, with the testing sample size set to 1,138, for training sample sizes from 1,138 up to 33,011, overall both gPRS and sPRS successfully predicted the brain phenotypes with significant correlations observed in the testing set, and higher accuracies noted for larger training sets. In addition, gPRS outperformed sPRS by showing significantly higher prediction accuracies across 13 brain phenotypes, with greater improvement noted for training sample sizes below ∼15,000. These findings support that GRE may serve as the primary genetic variable in brain phenotype association and prediction studies. Future imaging genetic studies may consider GRE as an option depending on the available sample size.
BibTeX
@misc{chen2023imputed,
  author = {Jiayu Chen and Zening Fu and Armin Iraji and Vince D. Calhoun and Jingyu Liu},
  title = {{Imputed Gene Expression versus Single Nucleotide Polymorphism in Predicting Gray Matter Phenotypes}},
  year = {2023},
  doi = {10.1101/2023.05.05.23289592},
  howpublished = {medRxiv},
}
Figure