Allelic imbalance in apeglm
1
0
Entering edit mode
Vincent.C. • 0
@vincentc-22254
Last seen 5 months ago
Laval

Hello,

Can the apeglm software be used to model (and test) the allelic imbalance (AI) as well as its shift between conditions? In the F1000 article, it is stipulated that estimating the intercept "in an intercept only model' is equivalent to the overall Allelic imbalance, which would suggest in a normal scenario using all sample of a similar nature. What I am wondering more specifically, is to assess the AI overall as well using samples with different condition, would the intercept remain an equivalent approximation of the basic allelic imbalance? If not, could one proceed to get this information using all sample (for a better model) as well as the in between condition shift using apeglm? Thank you !

apeglm Allelic imbalance concept • 328 views
0
Entering edit mode

2
Entering edit mode
@mikelove
Last seen 5 hours ago
United States

You could use a model matrix with contrast coding, eg an intercept and then -1/2 and 1/2 to specify differences between two groups for example. Then the intercept will still correspond to the overall allelic ratio.

0
Entering edit mode

Hi, Thank you for your response, just to make sure i am understanding you correctly. The matrix would be :

conditions<-c(-1/2,-1/2,-1/2,1/2,1/2,1/2)

x <- model.matrix(~conditions)

Would a continuous variable be used in this scenario by centring it around 0 (from 1/2 to -1/2) ?

1
Entering edit mode

You can center a continuous variable, yes.

You could do: x.center <- (x - mean(x)) / sd(x) so the LFC is more interpretable (log2 fold increase for 1 SD of x).

0
Entering edit mode

Hi,

Regarding this example, without a contrasting model, is it correct that the intercept estimates the Allelic imbalance of the levelled condition? So, by estimating each conditions' levelled intercept one could identify whether the shift emanated from (or resulted in) an imbalanced state?

Thank you for a very useful software and its support.

1
Entering edit mode

Yes that sounds right, the intercept gives the allelic ratio in the reference group (after logit transform).

0
Entering edit mode

Would in this case the logit function be necessary or would the betabinCR implementation be an adequate substitute ?

1
Entering edit mode

I donâ€™t follow. The coefficients returned by apeglm are not bounded between 0 and 1. You have to logit transform to get a ratio from a sum of coefficients. The method argument is not relevant.

0
Entering edit mode

I understand, I confused the upstream logit* method with this downstream reconversion from coefficient to ratio bound.

Thank you

*sorry log.lik

0
Entering edit mode

Got it. Good luck with the analysis!