Question

PCA from TPM

1

Entering edit mode

tanyabioinfo ▴ 20

@tanyabioinfo-14091

Last seen 5.7 years ago

Hi

I am trying to do PCA analysis of my samples. I generated the matrix using the tximport package. I have transcript ids as my rows and the sample names are the columns.

txi <- tximport(files, type="salmon", tx2gene=NULL, ignoreTxVersion=TRUE,dropInfReps=TRUE,txOut = TRUE)
tpm <- (txi$abundance[apply(txi$abundance, MARGIN = 1, FUN = function(x) sd(x) != 0),])

tpm = log2(tpm + 1)
tpm_centered <- t(tpm-rowMeans(tpm))

pca = prcomp(tpm_centered , scale=TRUE, center=TRUE)

cols <- as.factor(as.numeric(colnames(tpm_centered)))

plot(pca$x[,1],pca$x[,2], xlab = "PC1", ylab = "PC2",main ="PCA replicate1", col =cols)

text(pca$x[,1],pca$x[,2], row.names(pca$x), cex=0.5, pos=3)

I have couple of question.

1. Is generating PCA plot from txi$abundance a good idea to plot the PCA

2. I am unable to get the output colored based on samples.

Can someone please help me

Thanks

Tanya

pca tpm tximport • 5.0k views

ADD COMMENT • link 7.1 years ago tanyabioinfo ▴ 20

0

Entering edit mode

Hi Michael

I am now using the follwoing code:

txi <- tximport(files, type="salmon", tx2gene=tx2gene, ignoreTxVersion=TRUE,dropInfReps=TRUE)
sampleTable <- data.frame(condition =samples$condition,time=factor(samples$time))
rownames(sampleTable) <- colnames(txi$counts)
dds <- DESeqDataSetFromTximport(txi, sampleTable, ~condition+time)
dds <- dds[ rowSums(counts(dds)) > 1, ]
rld <- rlog(dds, blind = FALSE)
plotPCA(rld, intgroup = c("condition", "time"))

I have wild type and mutant as condition and the time point as 0hr 6hr and 24 hr. I have 4 replicate for each one of them. In the PCA plot the replicates are not grouping together. Do you think this is normal or I am making some mistake.

Regards

Tanya

ADD REPLY • link 7.1 years ago tanyabioinfo ▴ 20

0

Entering edit mode

Make sure that files and sample table are the same order. This is very important.

ADD REPLY • link 7.1 years ago Michael Love 42k

score 0 · Answer 1 · 2017-10-26

I would recommend generating PCA plot from the normalized transformed counts. There are statistical reasons to prefer variance stabilized measurements, and the normalization takes care of any biases that were corrected by the quantification software, as well as sequencing depth.

This would look like:

dds <- DESeqDataSetFromTximport(txi)
vsd <- vst(dds)
plotPCA(vsd)