Skip to content

DeGAs / dPRS

Summary

DeGAs (Decomposition of Genetic Associations) factorizes a large SNP-by-trait genetic association matrix into a small number of latent genetic components via truncated singular value decomposition (TSVD). The resulting components are used to compute the DeGAs polygenic risk score (dPRS), which predicts a target trait comparably to a standard genome-wide PRS while additionally attributing each individual's risk to specific, interpretable components — surfacing "outlier" individuals whose risk is driven by only one component rather than the typical multi-component blend.

Method

DeGAs performs truncated SVD on a matrix of GWAS summary statistics spanning 977 traits over 469,341 independent common variants, all drawn from the UK Biobank.[1] Each resulting latent component is a linear combination of variants with a characteristic pattern of trait associations. The dPRS for trait j is computed as a weighted sum of an individual's component scores, dPRS_j = U_j,* Σ Vᵀ G, where G is the individual's genotype vector, letting a person's overall genetic risk for a trait be decomposed into contributions from each component rather than reported as a single opaque number.[1]

Results

Analyzing 337,151 white British UK Biobank participants (training/validation/test) with replication in 25,486 non-British white participants, the authors showed dPRS performs comparably to a standard PRS while offering greater interpretability:[1]

  • BMI: genetic risk factorized into a fat-free-mass component and a fat-mass component. Most high-risk individuals had contributions from both, but a subset of "outlier" individuals showed strong loading on only one component — a distinction invisible to a single aggregate PRS.[1]
  • Myocardial infarction: risk components related to lung function, cholesterol, and blood-pressure medication use; five distinct genetic subtypes were identified among high-risk outlier individuals.[1]
  • The paper also examined gout as a third worked example (per its abstract), demonstrating the same component-decomposition-then-attribution pattern.

See Also

  • Individual-First Polygenic Risk Score Clustering — DeGAs profiles individuals across latent components rather than computing one genome-wide score, in the same spirit as this approach family
  • Genomic SEM — a related but distinct multivariate decomposition approach that models a small number of user-specified latent factors via structural equation modeling, rather than an unsupervised SVD over hundreds of traits

Citations

[1] Aguirre, M., Tanigawa, Y., Venkataraman, G. R., Tibshirani, R., Hastie, T., & Rivas, M. A. (2021). Polygenic risk modeling with latent trait-related genetic components. European Journal of Human Genetics, 29, 1051–1061. DOI: 10.1038/s41431-021-00813-0. Source: aguirre2021-degas.md. Supports: method description and all results above. Location: Full text — Methods and Results.