Combine polyA selection and ribodepletion experiments for clustering

[Jane M]

Hello everybody,

I have two RNASeq experiments that I would like to combine for clustering purpose. Unfortunately, my 2 experiments are rather different!

Experiment 1:
19 primary tumors, 2 conditions: 13 and 6 tumors per condition
Paired-ends, 75bp, Next Seq 500
TruSeq Stranded mRNA

Experiment 2:
12 tumor cell lines, the two same conditions: 6 and 6 cell lines per condition
Paired-ends, 100bp, Illumina HiSeq4000
TruSeq total RNA Stranded


I need to merge these two experiments since 4 of the cell lines derived from 4 primary tumors. The question is: Will a tumor and its cell line cluster together?

When I perform an ascending hierarchical clustering on the sample-to-sample distances with DESeq2 after merging the samples of the two experiments into one study, my samples cluster by library type.
It is not surprising since a lot of genes will have an expression value when using ribodepletion whereas they won't be captured by polyA selection.

I was thinking about simply eliminating genes that have 0 counts in polyA experiment and counts beyond a certain threshold for at least one sample of the ribodepletion experiment.

Do some of you had such problem and eventually a nice solution for exploiting such data?
Thank you in advance,
Jane

[Jane M]

I wonder if I should exclude genes not captured by both libraries before sizeFactors estimation: when looking at the log2 of raw and normalized reads counts for the samples of the two experiments, I can still distinguish the two experiments. Removing genes with different expression capture might improve sample normalisation, doesn't it?

After running some tests, it seems that there is no less expressed genes in the experiment with polyA than in the experiment with ribodepletion. I expected the contrary.

For example, 7360 genes have normalized counts below 20 for each 19 samples (polyA) and 9027 genes have normalized counts below 20 for each 12 samples (ribo).
6047 genes have raw counts below 10 for each 19 samples (polyA) and 4473 genes have raw counts below 10 for each 12 samples (ribo).

The "ribo" experiment was a bit more sequenced: 77 10^6 raw reads (60-102) for polyA vs 93 10^6 raw reads (73-130) for ribodepletion. The number of reads falling into exons of genes is similar in both experiments: 44 10^6 (30-56) vs 45 10^6 (33-70).

Any hint? Do I make a mistake somewhere?
Any feedback would be appreciated.