Is chimeric reads a problem in de novo assembly?

[lamz138138]

Hi, dear all!

I want to perform de novo assembly with four libraries (insert size were 270bp, 500bp, 2K and 5K), and the read are paired and the length is 150bp. After mapping the reads to reference with BWA, there are about 1/3 reads were chimeric for the two mate-pair libraries (2K and 5K). I don't know whether I should filter out these reads? There are little information after google. Considering the short libraries were used for constructing contig, then the reads from long libraries are mapped to contigs to link these contigs, in my opinion, the assembly tool still could use chimeric reads to link the contigs. However, my mate think there were rare chimeric reads in previous experiment since reads were short, and the assemble tool may can't deal with chimeric reads. Furthermore, I think if I filter out these reads, then this wouldn't be a true de novo assemble. So should I filter out the chimeric reads in mate-pair library?

Also, because I want to test ALLPATHs-LG, so I designed the 270bp library, but according to paper titled "Genome sequencing and comparison of two nonhuman primate animal models, the cynomolgus and Chinese rhesus macaques", the overlap reads were filter out when using SOAPdenovo, should I filter out these reads to use SOAPdenovo?

Any suggestion would be grateful!

Best wishes!
lamz138138

[GenoMax]

Are you sure the reads are chimeric? How did you make that judgement? MP library data requires special handling.

[lamz138138]

Hi, GenoMax!

I think this reads are chemical reads after mapping reads to reference (BWA mem). Take one pair read as example, read1 have two hit that one part of reads were mapped to position A and another part were mapped to position B, and read2 would mapped to position B. Besides, I have confirm the mapping result in browser.

Considering the re-fragment of circularized molecules is about 250~500bp, and the read is 150bp long, it is easy to produce chimeric reads.

[HESmith]

Mate-pair libraries are designed to be "chimeric", in the sense that non-contiguous genomic sequences become contiguous (and inverted) during library prep. The link provided by GenoMax illustrates this point. ALLPATHS-LG actually expects the mate-pair reads to have this type of structure.

But, given that you have a reference genome, what's the rationale for performing de novo vs. reference-guided assembly?

[lamz138138]

Hi, HESmith!

In my opinion, compare to reference-guide, de novo assembly may provide clue for structure variation.

In fact, I mapped the reads to reference to confirm the company had gave us the right data, and the experiment of mate-pair was successful. Then I found so many chimeric reads (the ratio is about 5% in the paper of SOAPdenovo), and considering the reads is longer than previously, my mate think the assembly tool couldn't deal with this type of reads, while I think it wouldn't be problem.

According to ALLPATHS-LG manual, I can only find it need overlap reads in short library, are you sure that it expect mate-pair reads in chimerica?

Thanks for reply!

[HESmith]

From the ALLPATHS-LG manual:

"Reads from jumping libraries may be chimeric, that is, they may cross the junction point between the two ends of the insert that occurs in libraries produced using the Illumina sheared library protocol."

N.B.-Jumping libraries = mate-pair libraries

[lamz138138]

Hi, HESmith!

Thanks very much, I got it in manual, why I had missed it......

Do you have experiment with SOAPdenovo, do you think it can deal with chimerica reads too. And should I use 270bp library to construct contig?