A few months back, when I asked Michael Schatz about the error rate of Oxford Nanopore, he gave me a very cryptic answer (‘assembly is possible, as long as there is more signal than noise’ - paraphrased). Thankfully, his newly posted biorxiv paper addresses the issue properly.
Major points -
Of the 267,768 reads produced by our 30 sequencing runs, 64,849 reads (24%) aligned to the reference yeast genome, and 31,013 (11%) aligned to the known spike-in sequence used for calibrating the instruments. The remaining 65% of reads did not show significant similarity to the W303 genome or spike-in sequence, presumably because of insufficient read quality (Supplemental note 4).
[Note: The comparative numbers for Pacbio are over 95%. This is very important for de novo assembly from long reads alone, because one does not know which reads are real and which are not.]
Supplemental Figure S5A shows that the mean identity to the reference of 1D reads was 64% while 2D reads produced many reads exceeding 75% identity.
[**Note: The comparative number for Pacbio is 85% identity. This is important for de novo assembly from long reads alone, because when one aligns long noisy reads with 85% accuracy against long noisy reads with 85% accuracy, the relative similarity becomes 70% (Edited, thanks Cedric). For 75% accuracy of reads, the relative difference will be 50% !! **]
Of the reads that align, the overall coverage distribution approximated a Poisson distribution centered at 35x coverage, although some over-dispersion was observed that was better modeled by a Negative Binomial distribution (Supplemental Figure S5C). To examine some of the sources of the over- dispersion we also examined the coverage as a function of the GC composition of the genome. Between 20% and 60% GC content, the coverage was fairly uniform, while at high and low GC content the coverage is more variable partially explaining some of the regions of the genome lacking raw read coverage (Supplemental Figure S5D).
[Note: Pacbio reads do not show GC bias. This is important for assembling Plasmodium and other genomes, which are very AT-rich. Also, it is important for assembling the AT-rich parts of the genome.]
The very longest reads tend not to be alignable at all, suggesting that the longest reads may be extremely low quality or include other artifacts of the sequencing process.
[Note: Very longest Pacbio reads are not trustworthy either.]
The remaining paper is not very useful. I have two major objections -
(i) You have to look very hard to find out that they are doing hybrid assembly and not a straight nanopore-only assembly. Why is it so difficult to mention that simple fact in the title/abstract?
(ii) If they are doing long-short assembly of yeast genome anyway, why stops them from doing a comparison with Pacbio-Illumina hybrid assembly and mention the comparative results in the abstract?