Combine full-length and 3` RNAseq, is it possible?

Question

[garcesj]

Hi there,

I'm faceting a vital dilemma and I'd need some advices, please. Up to now, I've processed some samples according a 3`-based RNAseq protocol... but currently I have the option to process the new ones with a full-length protocol (that will give me, theoretically, more information).

I guess, according the paper I attached, it's not feasible (or correct) to directly compare the final matrix counts... so I should realign my BAMs to a custom reference containing only the 3' ends for each gene. Do you know if there's already any way to do this? Or there is any study that have already done this? (I've found nothing).

Beyond technical aspects, what's your view about using two different (very different) protocols? Maybe could be better to use the same for the entire project?

Thanks a a lot. Bests.

[benformatics]

To clarify, it can be done you could attempt batch effect correction assuming you have comparable time points/experimental replicates. Would I recommend it? Absolutely not, as others mention below the headache involved in de-convoluting the technical effects from real biological effects would be not worth it at all. A major question is: why do you suddenly want more information? If you are simply repeating the same experiment with full-length transcript information to investigate alternative splicing or alternative promoter usage then analyzing the two datasets separately (3' vs full-length) and then comparing them is totally OK. Merging them together for analysis like DEG would not be fun.

[garcesj]

Thanks for your reply. How would you compare two different analysis?

Answer 1

[ATpoint]

Definitely use the same within the same project and make sure all batches you ever produce and plan to analyse together have replicates of all involved experimental groups to avoid confounding. Using different kits for the same running project is one of the worst sins in experimental design I could think of. This is nothing that can be corrected in silico, unless you have like half of the samples with kit A, and the second half with kit B, with the above mentioned replicates of all groups in both "batches". Even then it is suboptimal, don't do it. Kit is a major confounder in any NGS experiment.

Answer 2

[MohammadAlkadi]

In my humbel exprience, any difference in the protocols or computional method would result in bias in the count. I don't recommend it.