omerwe / polyfun

PolyFun (POLYgenic FUNctionally-informed fine-mapping)
MIT License
96 stars 22 forks source link

HESS estimates and zero heritability #114

Closed freg1086 closed 2 years ago

freg1086 commented 2 years ago

Hey Omer, it’s me again.

Just ran into a strange error when launching fine-mapping with Susie.

For a few regions/jobs, the Susie output looks like :

[INFO]  Average local SNP heritability estimated by modified HESS over 100 iterations: -4.5357e-04
Traceback (most recent call last):
     raise ValueError('HESS estimates that the locus causally explains zero heritability')
ValueError: HESS estimates that the locus causally explains zero heritability

I’m using the -hess option, so I guess the problem comes from me not setting any --hess-min-h2 value…any recommendations for that ? I’m working with UKB data (N=337K).

Thanks in advance!

omerwe commented 2 years ago

@freg1086 that's a good question! Unfortunately there's no clear-cut way to specify --hess-min-h2. Please see the following discussions for some details (including recommendations specific to UKB): #56 #70 #71

Hope this helps, please let me know if not!