Hi, I’ve run my paired samples as both unstranded and reverse stranded. Surprisingly I got very similar assigned reads in featurecounts (aroun 70%).
I’ve run the infer experiment, and results show it’s actually reverse stranded.
I was wondering how could this happen?
This is how the data actually aligned, and you could rerun any steps where the strand assignment was set differently. You’ll likely get “better” results if those steps are filtering or differently labeling data based on strand.
- Featurecounts does consider strand.
- If you used Stringtie upstream, that can also consider strand-specific data but only if you input annotation at that step and specify how to use it (only report known, or to report all).
The strand in this tutorial is set incorrectly, and is pending a fix, but with the tools/parameters that are currently in place the result is about the same whether strand is specified correctly or not.
Why? Stringie will create “novel” transcripts for everything, then merge with “known” annotation during Stringtie merge. Any transcript/gene that doesn’t end up with a known attribute can keep the novel attributes. This masks the strand “problem”, and is a very good example for how things can go wrong scientifically even when tools don’t error for technical reasons.
When running analysis with your own data – investigate the strand. Even if whoever published that data states that it is of a particular type, double check 
Good catch!