| Title: | Polygenic modeling of genetic effects on both phenotypic mean and variance: distributional regression for BMI, blood and urine biomarkers in the UK Biobank |
| Journal: | Frontiers in Bioinformatics |
| Published: | 11 Jun 2026 |
| Pubmed: | https://pubmed.ncbi.nlm.nih.gov/42369766/ |
| DOI: | https://doi.org/10.3389/fbinf.2026.1800403 |
Publication 20776
WARNING: the interactive features of this website use CSS3, which your browser does not support. To use the full features of this website, please update your browser.
Abstract
Polygenic scores (PGS) are commonly used to estimate the cumulative genetic contribution to complex traits by aggregating the effects of multiple genetic variants. In the case of continuous phenotypes, traditional methods have primarily focused on predicting the effect of variants on the phenotypic mean, thereby overlooking potential genetic influences that modulate phenotypic variance. Recent studies suggest that variance quantitative trait loci (vQTLs) provide important insights into the genetic control of trait variability and may serve as candidates for gene-environment interactions that are not captured by mean-based models. Using UK Biobank data, we applied snpboostlss, a cyclical gradient-boosting framework for Gaussian location-scale models, to derive sparse polygenic models for both the mean and the variance of quantitative traits simultaneously. We analyzed BMI and 30 blood and urine biomarkers (e.g., cholesterol, glucose, phosphate, urate) where joint genetic, environmental, and lifestyle contributions (e.g., sedentary behavior, diet) are expected. The distributional regression approach efficiently processes large-scale and high-dimensional genotype data by selecting, in each boosting iteration, a batch of variants that exhibit strong correlations with the current residuals. By estimating genetic effects on both the mean and the variance of quantitative traits, snpboostlss allows the construction of dual-component polygenic models that offer a more detailed view of the genetic architecture underlying the trait. Our analyses revealed genetic loci that influence the level of the trait alongside markers that predominantly affect trait variability. Notably, some variants contributed to both components, while others were specific to variance, suggesting distinct mechanisms and potential evidence for gene-environment interplay. These findings demonstrate that integrating variance effects into polygenic modeling via distributional regression can improve model interpretability and yield refined predictive insights for complex traits within precision medicine. Also, by including a variance component, our model has the potential to stratify individuals who could particularly benefit from environmental changes and lifestyle interventions.</p>
6 Authors
- Kiran Kunwar
- Qiong Wu
- Hannah Klinkhammer
- Christian Staerk
- Andreas Mayr
- Carlo Maj
1 Application
| Application ID | Title |
|---|---|
| 135122 | Elucidating the genetic basis of human traits and diseases across the allelic spectrum |
Enabling scientific discoveries that improve human health