Scantwo - normal model for binary data

12 views
Skip to first unread message

Conor Simpson

unread,
Jul 29, 2021, 11:31:15 AMJul 29
to R/qtl discussion

Hi all,

I am having trouble with generating permutations using the binary model for scantwo. The model is not converging and does not seem to ever pass 1 permutation so trying to run 1000 seems impossible.

For all my binary traits, I get pretty much the same output using the binary or normal model - with exactly the same significant QTL being identified. Is this sufficient to justify using the normal model to generate scantwo permutations?

Thanks,
Conor

Karl Broman

unread,
Jul 29, 2021, 2:14:10 PMJul 29
to R/qtl discussion
I would only use permutation results that were derived with the same method I used for the analysis.

karl

Conor Simpson

unread,
Jul 31, 2021, 3:18:03 AMJul 31
to R/qtl discussion
Hi Karl,

I understand that but what I want to know is if it is OK to use the normal method for a binary trait since I cannot seem to generate scantwo permutations using binary method. The binary method for such traits yield the same results for scanone as it does when I use normal on these binary traits. So can I assume that it will be the same for scantwo and simply use the normal method even though they are binary traits?

Thanks,
Conor

Karl Broman

unread,
Aug 2, 2021, 9:30:02 AMAug 2
to R/qtl discussion
The binary model uses logistic regression, so there will be three main sources for differences in the results:
- forcing fitted values to be in the interval [0,1]
- handling differences in variance (if probability is near 0.5, the results will be more variable than if the probability is near 0 or 1)
- interactions will be on a different scale

karl

Conor Simpson

unread,
Aug 4, 2021, 5:11:39 AMAug 4
to R/qtl discussion
Thank you for this explanation!
Best,
Conor
Reply all
Reply to author
Forward
0 new messages