>> To remind you, when testing a disease against two closely-linked
>>(to each other) markers, using MLINK, I got:
>> 1 (disease), 2 lod = 4.7
>> 1, 3 lod = 2.7
>> 1, 2, 3 lod = 16!
>> The consensus was (which makes perfect sense - I was simply ignorant)
>>that the third value is "Ott's generalized lod score", which is based
>>on the probability of ALL THREE LOCI BEING LINKED versus ALL 3 BEING
>>UNLINKED. And I know that 2 of them are linked.
The consensus here seems to have been quite wrong then, and you weren't
ignorant, it's MLINK that's misleading.
The short and simple answer is that when doing multipoints MLINK doesn't
output a lod score (log base 10). Instead it outputs twice the natural log of
the likelihood ratio, which is an anachronistic statistic called a location
score. It differs from the lod score by 2ln(10), about 4.6. All talk of
testing the wrong intervals, standardising on the wrong likelihood or getting
"Ott's generalised lod score" is really a side issue. The real problem is that
MLINK doesn't output what it says it outputs as a "log like difference", it
outputs a 2ln(like) difference. My advice is to look at the raw log
likelihoods you get and ignore the totals and what MLINK claims are the
differences.
