GeneCounts comparable when mapped with different STAR overhang index

[Bono Cheong]

Hi Alex,

I apologise in advance if I had submitted a post earlier about this issue. I'm not sure if it had went through, but here is the situation that I am having, which I don't think is specified clearly from my other post earlier today.

So I have this very large RNA-seq dataset which contains hundreds of samples of varying read lengths (49, 50, 76, 100 and 125bp). I would like to analyse these samples together, particularly using the GeneCounts output after STAR mapping. To do so, I generated STAR indices for each of the cohorts, using the ideal overhang (48, 49, 75, 99, 124) respectively.

I was wondering if this approach is more accurate than just using the default overhang of 100 across all samples in my dataset? I would like to ask this, because based on previous posts, it seems that both methods are doable, but I am not sure if there is a huge setback in how I have approached this.

Thanks for the help!

[Alexander Dobin]

Hi @bonocheongstandard

the approach with ideal overhang should be (very slightly) more accurate.

However, in most situations, the gains would be negligible.

[V J]

I have similar case; where a large dataset (~1800 paired read samples) with 151bp was analyzed with overhang 100. The ideal overhang is 150 quite different from the value used i.e. 100. Would your recommendation of " negligible gains in most situations" still hold? if using ideal parameter 150 is used to re-analyze the data.  The reason, I am posting is that redoing the whole analysis would require considerable time and resources and trying to gauge if it would be worth it. I appreciate your help!

[Alexander Dobin]

Typically, it's not worth it. You may want to check for a few samples if this is true in your case.