1. Is it possible to combine de novo and reference-guided assemblies?
Yes, you can combine de novo and reference-guided approaches. I've done exactly that with a reference genome since we couldn't afford optical mapping, Hi-C, or longer reads to better scaffold at the time. I realise you asked about transcriptome and not genome, but I can't see why this wouldn't work.
4. What are the benefits and potential disadvantage of doing so?
There are a lot of factors to consider, but first here some questions:
Why aren't you using the existing published reference genome?
How good is your data?
What improvements does your data offer over the existing published data?
The benefits could be moving towards a pangenome approach, where you take into account lots of variation. Downsides include the time investment, inheriting errors from the other assembly, adding new errors because you couldn't resolve evolutionary differences between assemblies. I'm sure there are more, but those are just off the top of my head.
Hopefully someone comes along with tools and workflow, it's been a few years since I worked on something like this so I am definitely behind the field.
