Scaffolding a genome with hybrid data

Question

I am assembling a ~500MB genome, and have ~150x long reads and ~200x 150bp PE short reads, with a ~400bp insert size.

I've done a lot of work with minimap+miniasm, and have what I think is a good set of unitigs and a .gfa file from miniasm. Depending on the parameters, I have a few hundred to ~1K unitigs. I can easily polish these unitigs with pilon, racon, and nanopolish, so that they are as error free as possible.

The question I have is: given access to

1. Polished (or not) unitigs
2. 150x long reads (long enough to span almost all repeats in the genome)
3. 200x 150bp PE short reads (~400bp insert size)

What would you do to scaffold a genome? And what are the pros and cons of your suggested approach?

What I need here is a draft genome. I would prefer a conservative approach that is more fragmented and has fewer misassemblies. So of course one option is just to keep the unitigs and make no attempt at scaffolding. But it does strike me there may be suitably conservative approaches I haven't considered. E.g. perhaps I could use ABYSS to scaffold my polished unitigs using my short-read data, thereby leveraging whatever information I can from the PE short reads?

Answer 1

Try LR_Gapcloser.

I've used L_RNA_Scaffolder for trying to scaffold a genome (which turned out to be more complex than I had expected). It looks like LR_Gapcloser (written by the same people) is similar, but designed specifically for scaffolding using long-reads.

That page also suggests PBJelly and GMCloser as competing tools.