The limma package

Differential Expression Analysis with limma in R

John Blischak

Instructor

Advantages of the limma package

Testing thousands of genes would require lots of boiler plate code

pval <- numeric(length = nrow(x))
r2 <- numeric(length = nrow(x))
for (i in 1:nrow(x)) {
  mod <- lm(x[i, ] ~ p[, "er"])
  result <- summary(mod)
  pval[i] <- result$coefficients[2, 4]
  r2[i] <- result$r.squared
}

Improved inference by sharing information across genes
Lots of functions for pre- and post-processing (see Ritchie et al., 2015 for an overview)

biocManager::install("limma")

Specifying a linear model

$$ Y = \beta_0 + \beta_1 X_1 + \epsilon $$

$Y$ - Expression level of gene
$B_0$ - Mean expression level in ER-negative
$B_1$ - Mean difference in expression level in ER-positive
$X_1$ - ER status: 0 = negative, 1 = positive
$\epsilon$ - Random noise

Specifying a linear model in R

model.matrix(~<explanatory>, data = <data frame>)

design <- model.matrix(~er, data = pData(eset))

head(design, 2)

      (Intercept) erpositive
VDX_3           1          0
VDX_5           1          1

colSums(design)

(Intercept)  erpositive 
        344         209

table(pData(eset)[, "er"])

negative positive 
     135      209

Testing with limma

library(limma)

# Fit the model
fit <- lmFit(eset, design)

# Calculate the t-statistics
fit <- eBayes(fit)

# Summarize results
results <- decideTests(fit[, "er"])
summary(results)

   erpositive
-1       6276
0       11003
1        5004

Let's practice!

Differential Expression Analysis with limma in R