I cultivated four different neuroblastoma cell lines and want to compare them with three tissue samples from healthy human brains to find the cell line closest to the brain. We used the single-color Agilent SurePrint G3 Human Gene Expression v3 8x60K Microarray Kit (P/N G4851C).

Let's name the cell lines CL1, CL2, CL3, and CL4 and the tissues T1, T2, and T3. I want to compare the following pairs:

• CL1 vs. T1
• CL2 vs. T1
• CL3 vs. T1
• CL4 vs. T1

and

• CL1 vs T2
• CL2 vs T2
• CL3 vs T2
• CL4 vs T2

and

• CL1 vs T3
• CL2 vs T3
• CL3 vs T3
• CL4 vs T3

I already imported the data to limma and background-corrected and normalized it. The resulting density plot looks good, so my data seems okay. From here, starting with my EList of background-corrected and normalized data, how can I run through my pairwise comparisons?

targets.tsv

FileName    CellType    Name    Cy3
file1.txt   CL1         CL1_1   Sample
file2.txt   CL2         CL2_1   Sample
file3.txt   CL4         CL4_1   Sample
file4.txt   CL3         CL3_1   Sample
file5.txt   CL1         CL1_2   Sample
file6.txt   CL3         CL3_2   Sample
file7.txt   CL2         CL2_2   Sample
file8.txt   CL4         CL4_2   Sample
file9.txt   CL1         CL1_3   Sample
file10.txt  CL2         CL2_3   Sample
file11.txt  CL4         CL4_3   Sample
file12.txt  T3          T3_1    Control
file13.txt  CL3         CL3_3   Sample
file14.txt  T3          T3_2    Control
file15.txt  T3          T3_3    Control
file16.txt  T1          T1_1    Control
file17.txt  T1          T1_2    Control
file18.txt  T1          T1_3    Control
file19.txt  T2          T2_1    Control
file20.txt  T2          T2_2    Control
file21.txt  T2          T2_3    Control
file22.txt  T3          T3_4    Control

script.R

# Targets
targetinfo <- readTargets(targetsFile, row.names = 'Name')

# Converts the raw data to an EListRaw object
wtAgilent.GFilter <- function(qta) { qta[,"gIsPosAndSignif"] }
project <- read.maimages(
  targetinfo,
  source = 'agilent.median',
  green.only = TRUE,
  path = "data",
  other.columns = 'gIsWellAboveBG',
  wt.fun = wtAgilent.GFilter
)

colnames(project) <- row.names(targetinfo)

# Background correction
project.bgcorrect <- backgroundCorrect(project, method = 'normexp')

# Normalization
project.bgcorrect.norm <- normalizeBetweenArrays(project.bgcorrect, method = 'quantile')

# Filtering
Control <- project.bgcorrect.norm$genes$ControlType != 0
NoSymbol <- is.na(project.bgcorrect.norm$genes$external_gene_name)
IsExpr <- rowSums(project.bgcorrect.norm$other$gIsWellAboveBG > 0) >= 3

project.bgcorrect.norm.filt <- project.bgcorrect.norm[!Control & !NoSymbol & IsExpr, ]

These are the relevant parts of my script. Now I wonder how to apply the linear fits to gain access to the desired pairwise comparisons as

cellType <- factor(targetinfo$CellType)

design <- model.matrix(~cellType)

fit <- lmFit(project.bgcorrect.norm.filt.mean, design)
fit <- eBayes(fit, trend = TRUE, robust = TRUE)
results <- decideTests(fit)

doesn't seem to be quite what I want.

Thank you very much for every suggestion for improvement and, most of all, for every hint in the right direction to solve my problem.

You just need to define contrasts corresponding to the comparisons you want to make. Just define

design <- model.matrix(~0+cellType)
colnames(design) <- levels(cellType)

and then form the contrasts you want using makeContrasts. See the limma User's Guide or see https://bioconductor.org/packages/release/workflows/vignettes/RNAseq123/inst/doc/designmatrices.html

There is no need for wt.fun = wtAgilent.GFilter in the limma pipeline. Low intensity probes are not the same as low quality probes and limma is perfectly capable of handling low intensity probes correctly.