Improve illumina short read assembly using PacBio long reads

Question

[shachigahoimbi]

I am trying to assemble goat genome (genome size=2.9 Gb) and I have goat genome sequencing data from short and long reads

1. short read data from Illumina (genome coverage ~37x).
2. long read data from PacBio (genome coverage ~1.5x)

I have assembled Illumina short reads using ABySS and SOAPdenovo and got best N50 1884 at K-mer of 41. I would like to improve short read assembly using PacBio long reads data. Because of the low coverage (1.5x genome coverage) of PacBio data, I am unable to decide which software would be best for the improvement of N50 using long reads.

I tried HybridSPADES for hybrid assembly of my short and long read data but it is giving issue regarding memory (out of memory).

Please let me know, how could I improve short read assembly using low coverage (~1,5 X coverage) long reads.

[lieven.sterck]

What was your input read length of the illumina data?

an optimal Kmer of 41 seems pretty low , what range did you evaluate?

Answer 1

[colindaven]

Maybe you can't, 1.5X actually means 0X for a good proportion of the genome.

Generally, you want 20X + Pacbio coverage to make a good assembly.

It might pay to use another better assembly - I think a goat is available - for orientating your short scaffolds.

Answer 2

[jean.elbers]

You could use your best Illumina assembly as input for whole-genome alignment with Cactus (https://github.com/ComparativeGenomicsToolkit/cactus) to NCBI accession GCA_004361675.1 as the reference.You could then use Ragout (https://github.com/fenderglass/Ragout) to generate a reference-guided assembly of your individual based off of the best available goat genome assembly GCA_004361675.1.

Answer 3

[lieven.sterck]

Since you're already on the ABySS route, you could give LINKS a try: that's a long read scaffolder from the same people/group as ABySS.

but as mentioned by others here as well, 1,5x coverage will likely not get you very far

Answer 4

[Vitis]

filtlong may help you filter and correct long reads using your short reads.

https://github.com/rrwick/Filtlong

Then the corrected long reads may help you scaffold some contigs. But I agree with the other answers: 1.5X of long reads wouldn't get you very far.