Improve fragmented genome assembly with PacBio?

[BasVerb]

I'm currently in the process of assembling a genome scaffold (estimated size 1-1.2 Gb). So far I have Illumina data available, paired-end and mate-pair (50x and 5x coverage estimated). Unfortunately the quality of the Mate Pairs is not great, which appears to make scaffolding problematic.
After assembly with SOAP-denovo2 I get on average around 3 million scaffolds (k 27-127), highly fragmented (perhaps the genome is repeat rich?). I'm trying the ALLPATHS assembler as well at the moment to see if that makes a difference.

To improve our scaffolding we were wondering whether it would be worth it to get some PacBio data, making use of the longer reads. Would this be a worthwile investment and if so, how much would be required (1 - 10x or more?), for budget reasons obviously. Or are there other things that can be done with our current data that might improve assembly?

Best wishes,

Bas

[sarvidsson]

You will need good mate pairs (or possibly PacBio, I have no experience there however) - what kind of quality problems do you have with yours? What kind are they (kit, targeted size) and how did you pre-process them?

I can recommend Allpaths-LG, when fed reasonable mate pair libraries, it is a very good assembler. However, if you have mate pairs with no clear insert peak, or data from a polymorphic population or highly repetetive genome, don't expect an assembler to perform magic...

[flxlex]

PacBio recommends 10x coverage in long reads for assembly improvement.

[Smurali]

Hi BasVerb,
my two cents on PacBio scaffolding:
PacBio reads can be used in different ways. If you have a high enough coverage, you can self-correct reads and use them for scaffolding (but this requires coverages >20x and is expensive).
However, if you are using raw PacBio reads, and scaffolding a very fragmented assembly (like in your case) then you can expect a lot of errors because of the read anchoring multiple different contigs. Make sure you watch out for this!
PacBio does well if you are improving genomes with fairly large contigs. Many of the primate assemblies in NCBI were (super)scaffolded with ~15x PacBio.

[Markiyan]

If possible get pacbio to 25X and do error correction before the assembly.

Dear BasVerb,

A couple of things.
Assuming it is not a highly polyploid genome:
Your assembly results look a bit too fragmented for the beast, so please do the following.
1. Make sure the adapters have been fully removed - do a fastqc run on the input data
(I assume that you had done your hiseq / miseq runs in 2x250 or 2x300 bp mode, if it is in 2x100 - rerun in 2x250 bp mode, preferably using PCR-free library prep for 350bp).
2. Preprocess reads with flash or panda. Do not use illumina nextera matepair linker adapter removal! If you have enougf overlapping PE matepair fragments - try using them, once split on the linker sequence (like 454 PE).
3. Screen for high coverage contaminants - bacterial genomes, mitochondrial DNA, etc - basically assemble a few small subsets, like 10K reads, 1M reads, and have a look at high coverage contigs, than use them as "vector" sequence when trying to assemble all data.

4. If you can get hold on the pacbio data - ask them to use the size selection (blue pippin or PFGE) to maximize the fraction of the library with 25kb fragments during library prep.

5. If possible get at least 25X PacBio coverage, but the exact results would be wery genome/DNA quality specific.

5. Do a first pass of the error correction on the pacbio reads before the main assembly using illumina data.

Try different assemblers. As a commercial draft - give a CLC bio a go, and see what comes out of it.

Markiyan.