plotCounts to visualize variations of a specific genes along conditions
1
1
Entering edit mode
Laia ▴ 10
@239caad3
Last seen 13 months ago
Belgium

Hi Michael,

Now that I could enter my full matrix of samples and I ran LRT for each factor in my design, I would like to see in which conditions are my genes of interest being up/downregulated.

I am doing this manually by plotting specific genes using plotCounts. But sometimes I see weird outliers that were not spotted in the initial PCA for all the samples.

This is an example; in this case, sample BN has very high counts for Gene X compared with the other red replicates in this condition

I am using the dds object to plot this:

d <- plotCounts(dds, gene="GeneX", intgroup=c("FactorSH", "FactorSTt"), returnData=TRUE)

ggplot(d, aes(x = FactorSH, y = count, color = FactorSTt)) +  geom_point(position=position_jitter(w = 0.1,h = 0)) +
geom_text_repel(aes(label = rownames(d)), max.overlaps = 5) + theme_bw() + ggtitle("GeneX") + 
theme(plot.title = element_text(hjust = 0.5))

If I get signif. results in the LRT test for this gene, but when I plot it I see this, should I discard that result? Or am I just missing one normalization step to visualize those counts?

Thank you.

Laia

DESeq2 plotCounts vst • 2.3k views
ADD COMMENT
0
Entering edit mode

I found out that BN is indeed an outlier (By plotting a hierarchical clustering heatmap).. so I think I'll have to remove this sample. But why wasn't this obvious in the PCAplot?

Also, I see that I have 4 samples (belonging to the same sequencing "batch4") that give lower counts in general compared to the other samples.

When I want to apply the batch correction by adding it into the model (~batch + factorSH + factorSTt + Interaction), I get an error because batches are not totally overlapping between conditions. Batches correspond to different sequencing days (samples that needed to be sequenced again after some technical issue).

So I tried a different approach: ComBat_seq, to get a new "adjusted counts_matrix". But then some clusters that were expected in the PCA (and that appeared initially) disappear after correction.

Is it possible that ComBat_seq is being too harsh? Can this be adjusted? In fact, I can only see a batch effect for batch4. Can this one be "prioritized" for correction?

Thank you.

Laia

ADD REPLY
0
Entering edit mode

If it is an outlier in a small number of genes, it won't be PC1 or PC2, also DESeq2 picks top genes by variance for PCA plot, you can change this in plotPCA.

We have a recommendation for correcting for batches, which is to use surrogate variables (svaseq) or "factors of unwanted variation" (RUVSeq), both of which have examples in the rnaseqGene workflow:

https://bioconductor.org/packages/release/workflows/vignettes/rnaseqGene/inst/doc/rnaseqGene.html#removing-hidden-batch-effects

ADD REPLY
0
Entering edit mode
swbarnes2 ★ 1.4k
@swbarnes2-14086
Last seen 1 day ago
San Diego

You can remove genes like that by filtering your dds object something like this: This will only keep genes where 5 samples have at least 10 counts.

keep <- rowSums(counts(dds) >= 10) >= 5
dds <- dds[keep,]

That sample might behave better after this kind of filtering.

ADD COMMENT
0
Entering edit mode

Yes recommend this type of filtering when you have outliers like this. This would go above DESeq()

ADD REPLY

Login before adding your answer.

Traffic: 587 users visited in the last hour
Help About
FAQ
Access RSS
API
Stats

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.

Powered by the version 2.3.6