PI_HAT and Z-values for relationships (PLINK IBD calculation)
1
3
Entering edit mode
4.5 years ago
Ram 36k

Hello everyone,

I've searched a whole lot on the web and there doesn't seem to be a popular strategy of tackling this question:

Given the PI_HAT, Z0, Z1 and Z2 values between 2 individuals, is it possible to narrow down on the possible relationship between the two, in cases where it's not obvious (PO/FS/UN)?

I'll elaborate.

The Zx values are fraction of relatedness measures, as in fraction sharing 0 copies, fraction sharing 1 copy and fraction sharing both copies between 2 individuals. Totally unrelated people would have Z0 ~ 1 and duplicated samples/identical twins would have Z2 ~ 1. PO, FS, UN and to an extent HS relationships have well defined values for all Zx as well as PI_HAT

However, once you get to 3rd degree relationships and beyond, (PI_HAT~0.125), it gets murky. At that point, even the detailed breakdown with the Zx values don't help us figure out what the exact nature of the relationship could be. It could be first cousins, or it could be grandparent-grandchild. Both have PI_HAT~0.125.

In this case, I wonder if we have a comprehensive listing of all possible relationships for each PI_HAT fraction (0.125, 0.0625, etc) that I could then combine with clinical information (such as age) and then take a call on the actual relationship?

The closest I've gotten is this table in Wikipedia (and other sources, but Wikipedia is quite good, ironically):

Any help would be appreciated, thank you!

--
Ram

plink PI_HAT IBD • 8.2k views
4
Entering edit mode
4.5 years ago

The PI_HAT/Z0/Z1/Z2 values produced by plink --genome are generally not accurate enough to justify that kind of relationship inference. I recommend looking at KING or PC-Relate for this purpose instead.

1
Entering edit mode

@chrchang523, I'm trying to figure out what panel of SNPs KING interrogates to calculate kinship, but this doesn't seem to be explicitly stated anywhere. Do you know what this panel is?

0
Entering edit mode

I've tried KING, and 99% of the time, the Kinship coefficient is 0.5 * plink's PI_HAT. I'll run KING again, but how can I use it to infer relationships?

EDIT: Never mind, I see the --related option. Thank you! RTFM, I know!