Relatedness issue - Plink 1.9 IBD vs. Plink 2 KING vs. GREML GRM

[Lindsay Melhuish]

Hi,

I was just wondering if you were aware of differences in how relatedness is calculated across the various programs? 

When I check my data using PLINK (PI_HAT score of 0.12 or 0.2) and KING kinship coefficient of 0.0844, I have no related individuals. 

When I tried to estimate SNP heritability using GREML, the GRM told me I had related individuals, even at cut-offs 0.08. 

I don't understand how GREML works so I'm not sure how the calculations differ between PLINK and GREML. Which do I trust more? 

All the best,

Lindsay

[Christopher Chang]

It's possible to do better, but I trust the KING coefficient over plink 1.x's IBD and GRM estimates.  The latter two assume a single homogeneous population and require good allele frequency estimates for that population; KING remains effective with multi-population datasets.