DBG construction from short and long reads

[SimonHegele]

Hello,

and thank you for this amazing tool. I am currently examining different methods of hybrid de novo transcriptome assembly.
I constructed various assemblies from mouse data and compared their results and the result of a Stringtie assembly with BUSCO and rnaQUAST. In terms of BUSCOs metrics and most of rnaQUASTs metrics RNA-bloom gave the best results so far.

However, the number of mismatches in the alignments of the transcripts to the reference genome is not significantly lower in the hybrid assembly compared to the long-read assembly (~2.34 per kb vs ~2.4 per kb). So I wondered if the long and short reads are in any way treated differently in the DBG graph construction. If not this might explain the small impact of the short reads.

Best,
simon

[kmnip]

Hello Simon,
If both short reads and long reads are used, the short reads only contribute to the de Bruijn graph and the k-mer multiplicities in the error correction step in long read assembly. So, RNA-Bloom most likely would be correcting mismatches and short indel errors, leaving the long indels untouched. I think I can possibly implement a more aggressive strategy to use only short-read k-mers only.
Ka Ming

[SimonHegele]

Thank you for your reply. I think it it is generally a good idea to have the DGB constructed both from long and short reads so the long reads get corrected even in regions with no short read coverage. However, for accuracy one should trust the short reads over the long reads. Maybe this could be implemented simply by having the short reads contributing more to the k-mer multiplicity by adding a factor.