I think de novo assembly is the more accuarate but you need for that:
-Huge amount of sequences (I have a lot of reads that were acutally suposed to cover only one time the genome… it actually cover only one third, so impossible to make it)
-It needs a lot of memory… for my third of genome, computer answered "I cannot let you do this: it would requier one week with one TB of ram memory"
So: drawbacks are: expensive, time consuming…
If you have an existing genome, you might put any read / sequence on it and it might be quicker and cheaper.
Maybe another way could be from initial reads:
make a blast db from it
blast the part of the genome you are interested it
from the alignments design primers and clone the region you are interested in.
If some regions are closed from one another try to clone between them…
Then, you might have a partial assembly of what you are interested in.
You can also map what ever you have on a genome then combine with what just sugest and feel few gaps with old fation sanger sequencing for getting a reliable assembly of your interested part.
Eventually, if what you what to assemble is repetitive parts, a rad-sequencing form a repetitive part with 2 differant restriction enzymes might provide easy to assemble sequencing.
If you are interested in unique parts, theres new methods making a loop of DNA allowing to target it.