log fold change from glm/poisson

[fpbarthel]

Hi,

I'm wondering how to derive the log fold change from the output of R's glm using poisson family (for RNA-seq data). From R's lm function for linear models (for log2 microarray data) the log fold looks like it is the coefficient (beta), but it looks like the glm output needs some transformation to obtain log fold change. Anyone can walk me through the commands to do this?

Thanks!

[dariober]

Hi,

I'm wondering how to derive the log fold change from the output of R's glm using poisson family (for RNA-seq data).

Hi- The coefficients from glm with Poisson family distribution by R default are presented as natural log since the link function is log. so you have to anti-log the coefficient by exponentiation to get fold change. E.g.:

y<- rep(c(10, 40), each= 5)
x<- rep(c('A', 'B'), each= 5)
(fit<- glm(y ~ x, family= poisson()))

Call: glm(formula = y ~ x, family = poisson())

Coefficients:
(Intercept) xB
2.303 1.386

Degrees of Freedom: 9 Total (i.e. Null); 8 Residual
Null Deviance: 96.37
Residual Deviance: -3.997e-14 AIC: 52.44


fc<- exp(fit$coefficients[2]) ## Antilog coef #2
log2(fc) ## Convert to more familiar log2 fold change:
2