Single Step Genomic Evaluation via Hybrid Relationship Matrices

33 concepts17 papers2 proposals

In quantitative genetics, the goal is to predict an individual's traits—whether it is the milk yield of a dairy cow, the climate resilience of a crop, or a human's risk for a complex disease—by understanding their genetic similarity to others.

Historically, this was done using a pedigree relationship matrix ($A$), which tracks expected genetic sharing based on family trees. For example, full siblings are expected to share exactly 50% of their DNA. However, as Mayra Gómez Carpio et al. point out in their work, this assumes a theoretical average and completely misses the randomness of Mendelian segregation—the biological lottery where one sibling might inherit completely different alleles than another.

Modern genomics solves this with a genomic relationship matrix ($G$), built from direct DNA markers like Single Nucleotide Polymorphisms (SNPs). The matrix $G$ captures realized relatedness. But genotyping every single individual in a massive population or historical lineage is prohibitively expensive or physically impossible. This leaves us with a fragmented dataset: a large population with deep pedigree records and phenotypic data, but only a small subset with genomic data.

The solution is Single-Step Genomic Evaluation (ssGBLUP), a framework that mathematically glues these two worlds together into a single, unified hybrid matrix.

The Mechanism: How Information Propagates

Instead of running separate analyses or discarding ungenotyped individuals, ssGBLUP constructs a unified hybrid relationship matrix (denoted as $H$). As explained by Stefano Biffani et al. in their evaluation of buffalo breeding, this matrix blends the pedigree matrix $A$ with the genomic matrix $G$ to create a single covariance structure.

To understand why it works, imagine partitioning your population into two groups: ungenotyped individuals ($u$) and genotyped individuals ($g$). The resulting hybrid matrix $H$ is structured as follows:

$$H = \begin{pmatrix} A_{uu} + A_{ug}A_{gg}^{-1}(G - A_{gg})A_{gg}^{-1}A_{gu} & A_{ug}A_{gg}^{-1}G \\ G A_{gg}^{-1}A_{gu} & G \end{pmatrix}$$

The intuition behind this matrix is elegant:

  1. The Genotyped Block: For individuals with DNA records, the model uses their true, realized genomic relationships ($G$).
  2. The Imputation/Propagation Mechanism: Look closely at the top-left block ($H_{uu}$), which dictates the relationships among ungenotyped individuals. It starts with their baseline pedigree relationships ($A_{uu}$) and then adds a correction term: $A_{ug}A_{gg}^{-1}(G - A_{gg})A_{gg}^{-1}A_{gu}$.

The core of this correction is $(G - A_{gg})$, which represents the difference between the realized genomic relationships of the genotyped animals and their expected pedigree relationships. The matrix multiplication acts as a bridge, propagating these observed genetic deviations from the genotyped subset back through the family tree to their ungenotyped relatives.

By defining this joint distribution, genomic information flows seamlessly across the entire population. An ungenotyped animal's breeding value or risk score is adjusted because its genotyped uncle or sibling happened to inherit specific, measured genetic variants.

Why It Works: Eliminating Bias and Maximizing Power

Prior to single-step methods, researchers relied on multi-step approaches: they estimated genomic effects within the genotyped group and then tried to combine those results with traditional pedigree evaluations. This fragmentation caused severe selection bias, because the genotyped animals are rarely a random sample; they are typically selected based on historical performance recorded only in the pedigree.

By integrating all phenotypic, pedigree, and genomic data into a single computational step, the model avoids this bias. As noted by Huatao Liu et al. in their study on Yorkshire pigs, this unified framework maximizes statistical power because it evaluates the entire population simultaneously, rather than discarding valuable phenotypic records from ungenotyped ancestors or contemporaries.

The practical impact of this information propagation is substantial:

  • Accuracy Gains: In empirical studies, Stefano Biffani et al. demonstrated that single-step evaluation boosted the accuracy of genetic merit predictions by 20% to 40% for critical economic traits compared to pedigree-only models.
  • Robustness: Large-scale simulations by Xiaochun Yan et al. confirmed that this hybrid approach consistently outperforms models relying solely on DNA markers or solely on genealogical records, providing a highly cost-effective roadmap for breeders (Yan et al.).

To make these complex workflows accessible, software packages like PyAGH, developed by Qamar Raza Qadri et al., automate the fast construction of these combined kinship matrices, allowing researchers to easily model additive, dominant, and epistatic effects across diverse omics layers (Qadri et al.).

From Livestock Breeding to Human Biobanks

While single-step evaluation was forged in the context of livestock breeding—transforming industries like dairy cattle (Peng Wang et al.), swine, and goats—the exact same mathematical intuition applies to human genetics and precision medicine.

Consider the challenge of calculating Polygenic Risk Scores (PRS) for complex diseases at a biobank scale. Genotyping millions of individuals is financially daunting. However, Buu Truong et al. demonstrated a striking result in their paper: by using a pedigree-based linear prediction model that integrates the observed phenotypes of ungenotyped first-degree relatives, they achieved the same prediction accuracy with a sample of just 5,000 genotyped individuals as a traditional study would achieve with 220,000 unrelated individuals. This 44-fold increase in efficiency occurs because the model leverages shared genetic and environmental variance through known familial links without requiring direct sequencing for every family member.

Similarly, this framework allows researchers to exploit massive, ungenotyped historical datasets. Iris Kalka et al. utilized multi-generational genealogical records from electronic health records to construct population-scale kinship matrices, successfully estimating the heritability of drug responses without needing direct DNA sequencing for the entire cohort (Kalka et al.).

Refining the Signal

The core challenge in building these hybrid matrices is ensuring mathematical consistency. Because pedigree matrices reflect theoretical expectations and genomic matrices reflect actual molecular counts, simply pasting them together can introduce scaling biases.

This problem is closely related to a broader insight in genome-wide association studies. As Christian Widmer showed in his research on linear mixed models, genetic prediction is optimized when we use a "mixture" of similarity matrices—combining a broad background matrix representing global ancestry with a targeted matrix derived from trait-predictive markers.

Single-step evaluation operates on a similar philosophy. By treating relatedness not as a static, single-source metric, but as a multi-layered composite of historical lineage and direct molecular observation, it allows geneticists to see through the noise of the genetic lottery and map the true flow of heritable traits.

Key Foundational Work

A novel approach for haplotype-based association analysis using family data (2010) — contributed 2 concept elements to this direction.
Bayesian Pedigree Analysis using Measure Factorization (2012) — contributed 1 concept elements to this direction.
History & citationVersion 2 · 2026-10-04

Current · Permanent link

  1. Version 2 (latest) · 2026-10-04 · 16 papers
  2. Version 1 · 2026-10-04 · 16 papers
Cite this direction

Latest version (a living page):

@misc{lacuna_single_step_genomic_evaluation_658cedad,
  title = {single-step genomic evaluation},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/single-step-genomic-evaluation/rdr_658cedad822a43ce9445702488a4b58a},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 2, frozen text):

@misc{lacuna_single_step_genomic_evaluation_658cedad_v2,
  title = {single-step genomic evaluation},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/single-step-genomic-evaluation/rdr_658cedad822a43ce9445702488a4b58a/v/2},
  note = {Version 2, frozen 2026-10-04. Accessed 2026-11-19},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake