Hi I have a question about DEXSeq batch effect design. I followed the protocol from DEXSeq guideline to count in the batch effect.

featureID <- as.factor(data $featureID)  
groupID <- as.factor(data $gene_symbol)
design <- formula( ~ sample + exon + condition:exon )
formulaFullModel = ~ sample + exon + type:exon + condition:exon
formulaReducedModel = ~ sample + exon + type:exon
sampleTable = data.frame(row.names = c(paste0("num_",sample_3[1]),paste0("num_",sample_3[1+1]), paste0("num_",sample_6[1]),paste0("num_",sample_6[1+1])), condition = c("Control","Control", "Treatment","Treatment"), type = c("S","M"))

dxseq = DEXSeqDataSet(data1, sampleTable, design, featureID, groupID)
dxseq = estimateSizeFactors(dxseq)
dxseq = estimateDispersions(dxseq, formula= formulaFullModel)
dxseq = testForDEU(dxseq, reducedModel = formulaReducedModel, fullModel = formulaFullModel )
dxseq = DEXSeqResults(dxseq)

I also did DEXSeq without counting the batch effect. Without batch effect I've got more number of significantly regulated isoform than with batch effect design. According to their guideline, I suppose to observe more significant cases with the code above. I tried this method with different dataset and I always has less significant cases with the batch effect formula.

Looking forward to hear any explanation to understand the reason.

Thanks! -A