SNP heritability: What are we estimating?

Abstract

The SNP heritability has become a central concept in the study of complex traits. Estimation of based on genomic variance components in a linear mixed model using restricted maximum likelihood has been widely adopted as the method of choice were individual level data are available. Empirical results have suggested that this approach is not robust if the population of interest departs from the assumed statistical model. Prolonged debate of the appropriate model choice has yielded a number of approaches to account for frequency- and linkage disequilibrium dependent genetic architectures. Here we analytically resolve the question of how these estimates relate to of the population from which samples are drawn. In particular, we show that the correct model for the purpose of inference about does not require knowledge of the true genetic architecture of a trait. More generally, our results provide a complete perspective of these class of estimators of , highlighting practical shortcomings of current practise. We illustrate our theoretical results using simulations and data from UK Biobank.

Evaluation and characterization of expression quantitative trait analysis methods in the Hybrid Rat Diversity Panel

“…The reader will note a discrepancy between the heritability estimate and displayed in Figure 2 and the heritability estimate displayed in Figure 3 , which is due to the following. The GEMMA heritability estimation described in Figure 3 represents the narrow-sense heritability, which is defined as the fraction of phenotypic variance explained by additive SNP effects ( Rawlik et al, 2020 ). Note that narrow-sense heritability estimation is naturally specific to the set of genotyped SNPs.…”

SNP and Haplotype Regional Heritability Mapping (SNHap-RHM): Joint Mapping of Common and Rare Variation Affecting Complex Traits

“…Additionally, it would be interesting to explore the incorporation of other GRMs that account for allele frequencies and LD ( Speed et al, 2020 ) in the genomic background correction stage of SNHap-RHM, going forward. However, whether that will perform better than the commonly employed standard GRM proposed by VanRaden (2008) remains unclear ( Rawlik et al, 2020 ). Moreover, for the regional matrices in SNHap-RHM, it is important to retain SNPs in LD as these determine the haplotype structure that we wish to explore.…”

SNP‐based heritability and selection analyses: Improved models and new results

“…If such correlations do exist then ĥ2 SNP based on (4) can be biased. [36][37][38] The effect-size distributions cited above imply the one-parameter uniform heritability model, 𝜎 2 j = h 2 SNP ∕m for all j. We note in passing that many SNP-based analysis and simulation models also assume that the expected heritability is uniform over SNPs, ignoring its actual dependence on known SNP properties such as MAF, LD with other SNPs, genotype quality and functional annotation.…”

“…If such correlations do exist then $$\hat h_{{\rm{SNP}}}^2$$ based on (4) can be biased. [ 36–38 ]…”

“…[ 44,45 ] In the limit of increasing n , K ′ produces unbiased estimates of $$h_{{\rm{SNP}}}^2$$. [ 38 ] Intuitively, K ′ reduces the impact on inferences of sets of correlated SNPs, similar to the aim of LDAK weights which are much less computationally demanding to derive than K ′.…”