Which is the purpose of the transcript assembly?. Do you have any genome reference? annotation?.

In most RNAseq analysis, if you just want to compare transcript expression between conditions (or stages), the best strategy may be just to perform a basic differential expression analysis (DE), there is no needing of assembly the reads.

If you do not have any genome or annotation references, just in that case may be useful to assembly all transcripts (all reads from all stages) and then use all these transcripts as a reference to perform the DE.

I agree with Francisco, it all depends on wether you have or not a reference transcriptome, in that case you could just align your data on the reference. I suggest you to use the "classic" HISAT2-Ballgown pipeline which is sort of a convention (https://www.nature.com/articles/nprot.2016.095). The article explains clearly every command.

If you don't have a reference you'll probably need to assemble all the transcriptomes, or just assemble one of them and use it as the reference (choice a "control" sample).

Check these: Article Usability of reference-free transcriptome assemblies for det...

https://www.nature.com/articles/s41598-019-44499-3

As proper answers were already posted above, I would like to add

• None of the approach you mentioned can be voted as best, unless you add information about your aim/references (first point from Francisco Callejas-Hernández )
• All approaches can be equally good, if you want to compare the expression irrespective of drawing hard conclusion
• In a more exploratory scenario, do all the analyses you mentioned, and interpret according to previous data/references and your hypothesis
• It might also be possible that by doing so or some trial and error, you can discover a better strategy/results/method which has not be applied before or published (in your case or in general)