Multiple QTL on the same chromosome or just one?

54 views
Skip to first unread message

Jason Johns

unread,
Oct 1, 2021, 7:02:37 PM10/1/21
to R/qtl discussion
Hi Karl,

I've inquired about this issue before, but I'm hoping I've narrowed my question down enough to make it manageable. I'm stumped with my situation on chr2 determining how many of these peaks are actually independent. Via my unconfident analysis, it seems like there may just be 1 peak on chr2 @ 40cM, although there could very well be at least one more. I am especially suspicious because we have a candidate gene on chr2, corroborated by functional data that falls under the most distal peak on chr2. Anyway, I would be very grateful to have your guidance.

Scanone and scantwo analyses identify 3 loci on chr2 that may be QTL (attached). 

When I consider the largest QTL (chr3) as a covariate, the most proximal peak on chr2 ('1' as I have it labeled) all but disappears and the LODs of the two more distal peaks on chr2 increase by ~50%. When I make chr2, peak '2' a covariate, the most proximal peak ('1') disappears and the most distal one ('3') is reduced. When I make chr2, peak '3' a covariate, the most proximal peak ('1') is reduced but still significant and peak 2 is barely significant.

Another approach I took was to build a multiple model, first just considering the middle peak ('2') on chr2, then running 'addqtl' and 'addpair'. 

Here's the addqtl with the chr2 locus included. It doesn't look like there's much solid evidence for an added QTl on chr2, although there seems to be weak evidence for a QTL on chr4.

> mod3_add <- addqtl(std_210928, qtl = mod3_refined, formula = y~Q1+Q2+Q3+Q4+Q5, pheno.col = "std")

> summary(mod3_add)
         chr    pos       lod
c1.loc43   1 43.0   3.096
c2.loc55   2 55.0   2.632
c3.loc46   3 46.0   2.595
c4.loc16   4 16.0   4.602
c5.loc40   5 40.0   2.882
6.28           6 39.9   2.312
7.015         7 20.1   0.833

Here's addpair without the chr2 locus included to see if 2 loci should be added to chr2. Again, not much evidence for 2 QTL on chr2 from what I can tell. 

mod3_ap <- addpair(std_210928, qtl = mod3_refined, chr = 2, formula = y~Q1+Q3+Q4+Q5, pheno.col = "std")
Scanning full model for chr 2 and 2 
Scanning add've model for chr 2 and 2 
> summary(mod3_ap)
      pos1f pos2f lod.full lod.fv1 lod.int     pos1a pos2a lod.add lod.av1
c2:c2    40    54     27.5    4.25    1.59        41    54    25.9    2.65

Thank you for any guidance you can give here.

Jason

std_scanone_211001.pdf
std_peak2a_control_211001.pdf
std_scantwo_211001.png
std_peak3_control_211001.pdf
std_peak2b_control_211001.pdf

Karl Broman

unread,
Oct 3, 2021, 12:24:21 PM10/3/21
to rqtl...@googlegroups.com
I would look at compare models with 1, 2, or 3 QTL on chr 2, at the best-estimated positions for a given number of chr 2 QTL, and then assess whether the increase in LOD, as you add QTL to the model, seems sufficient. 

karl

On Oct 1, 2021, at 6:02 PM, Jason Johns <jwjo...@gmail.com> wrote:

Hi Karl,
--
You received this message because you are subscribed to the Google Groups "R/qtl discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to rqtl-disc+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/rqtl-disc/4b245312-40d8-460f-a2d8-95d31b579c15n%40googlegroups.com.
<std_scanone_211001.pdf>
<std_peak2a_control_211001.pdf>
<std_scantwo_211001.png>
<std_peak3_control_211001.pdf>
<std_peak2b_control_211001.pdf>

Jason Johns

unread,
Oct 6, 2021, 2:57:54 PM10/6/21
to R/qtl discussion
Thank you very much for your response. A couple follow up questions: 

1. Is it less compelling to consider 1 QTL as a covariate and see how the LODs of other QTL change (as in the attached figure), as opposed to comparing overall LOD scores of models considering one, two, or three QTL?

2. In the case of chr2, I'm concerned that I have 1 QTL at the proximal end of the chromosome and 1 at the distal end. However, the program is considering the 1 QTL at the distal end as 2 distinct ones. So, when I build a multipleqtl model and input 2 QTL on chr2 (one proximal & one distal), refineqtl moves the proximal locus over toward the distal end (within 10cM of each other). I understand that the program is maximizing the LOD, but would it be worth following intuition on this and 'forcing' the model to have one proximal and one distal QTL on chr2? *This is of course, contingent on me actually being correct about my intuition.

As always, I'm very grateful for your guidance.

Jason
std_peak2b_control_211001.pdf

Karl Broman

unread,
Oct 6, 2021, 3:12:29 PM10/6/21
to rqtl...@googlegroups.com
As I understood it, you were trying to figure out which of three QTL were supported, by looking at the peaks in the single-QTL model or in the one-dimensional scan controlling for a QTL. I would instead ask: if I assume two QTL on this chromosome, where are they estimated to be, and is the fit of the two QTL model sufficiently improved over the single-QTL model? And then: if I assume there are three QTL on this chromosome, where are they estimated to be, and is the fit of the three QTL model sufficiently improved over the best three-QTL model?

I'm not sure how you decide that your intuition is to be preferred over what the data seem to be saying, unless you see that the data have some artifacts.

karl

Jason Johns

unread,
Oct 12, 2021, 1:18:42 AM10/12/21
to R/qtl discussion
Hi Karl,

I apologize for my poor communication. I was using the peaks in the single-QTL model, the one dimensional scan controlling for a QTL, and the results of the scantwo to feed multiple QTL models. Is it true that, given the imperfect data, the program could be treating what is actually 1 QTL as 2 separate loci? And one of the ways we diagnose this is by first, intuitively looking at their proximity and second, controlling for one of the peaks and seeing if the other diminishes? 

Then downstream, say we have determined that 2 QTL are actually the same locus and there is one other distinct QTL on the same chromosome. If the program still wants to consider what we determined to be one locus as 2 different QTL, and those are the two highest LODs on the chromosome, then refine QTL will consider those 2 highest QTL and ignore the 3rd that we believe to be distinct based on our upstream analyses. In this case, not until we consider 3 QTL on that chromosome will refineqtl incorporate the other locus on the chromosome. However, now we are considering 3 QTL when we have gathered evidence that there should be just two.

Am I totally off here, and should just be heavily relying on the multiple QTL model to determine which loci are worth including, without considering upstream analyses? If so, my understanding is that the general rule of thumb is a QTL 'sufficiently' improves the model if it increases the LOD score by >= the significance threshold (5%: 3.37 in my case)?

I'm really grateful for all of your help with this. If I ever meet you in person, all of your beers (or whatever you like to drink) are on me.

Jason

Karl Broman

unread,
Oct 12, 2021, 8:48:26 PM10/12/21
to R/qtl discussion
I guess I still don't quite understand the question, and I would prefer to focus on questions about the software rather than about analysis results. I prefer to stay away from questions like "are there three or two QTL, and if there are two QTL which two are the ones?"

karl
Reply all
Reply to author
Forward
0 new messages