Why limma output has logFC when I fit the model to a continuous variable?

Question

[repinementer]

I am trying to identify genes that are associated with age (upregulated or downregulate with the age). To do this I use the follwing code

# limma phenotype covariates
x.t$SMRIN = as.numeric(x.t$SMRIN)
x.t$Gender = as.factor(x.t$Gender_name)
x.t$SMTSISCH = as.numeric(x.t$SMTSISCH)
x.t$AGE = as.numeric(x.t$AGE)

design <- model.matrix (~x.t$AGE+x.t$Gender+x.t$SMRIN+x.t$SMTSISCH, data=x.t)

# fit the linear model
fit <- lmFit(mydata3, design)

#apply eBayes
fit <- eBayes(fit)

#print summary results
summary(decideTests(fit))

#sort by pvalue
Topmodel=topTable(fit,n=Inf,sort="p", coef=2)

The results look like this. What I don't understand is If age is a continuous variable, what does FC mean here? FC is usually between two conditions, correct?

# results
> head(Topmodel)
         logFC  AveExpr        t       P.Value     adj.P.Val        B
XIST  9.571102 9.725594 40.87398 8.479010e-123 6.918872e-120 269.9499
AST   8.953672 8.944053 39.29622 1.422760e-118 5.804862e-116 260.2891
TSA  8.635042 8.661460 38.05368 3.653929e-115 9.938687e-113 252.4883
LLBB 8.177704 8.255516 37.22488 7.561766e-113 1.542600e-110 247.1879
DDJ   8.406508 8.889198 37.03714 2.557951e-112 4.174577e-110 245.9764
KMS  8.410594 8.410444 36.86320 7.940392e-112 1.079893e-109 244.8502

[Kevin Blighe]

Did you search on Bioconductor forum, where this has likely been answered already? As far as I am aware, the fold change for a continuous variable can be somewhat interpreted as the beta coefficient ('estimate') from a linear regression fit. i.e., if the fold change is positive, it indicates that the continuous variable's values increase with your outcome. I don't know the exact calculations that are being used by limma, though.

Are you sure that you need all of those covariates in your design formula? XIST comes up likely due to gender-specific effects.

[h.mon]

Just to illustrate Kevin Blighe point:

Limma: Continuous variable model designs

[repinementer]

Thanks, Kevin. Yes, unfortunately, I couldn't find what I am looking for. I am correcting the effect of gender in aging-associated genes identification. Don't you think so? And also I changed the gene names intentionally.

Answer 1

[Gordon Smyth]

I am trying to identify genes that are associated with age

Your code looks ok for that purpose.

What I don't understand is If age is a continuous variable, what does FC mean here?

It is the log2-fold-change in expression that results from a unit change in Age. If Age is in years, then a unit change is 1 year.