How do sequencing errors modify de Bruijn graph connectivity? (Part II)

Continuing in the saga of "what do sequencing errors do to our de Bruijn graph density measure" (read the first post here), I have some new results.

The conclusion of the first post was that on random (non-real) genomes, both with and without repeats, we see that de Bruijn graph connectivity is decreased by random sequencing errors. Zam Iqbal and I had a reasonably robust discussion in the comments, and he suggested trying a real genome. (Yes, it was on my list. But he upped the ante by saying he didn't believe my results were relevant because they weren't real genomes. Fair 'nuff!)

So I applied the make-reads.py and make-reads-biased-random.py scripts to E. coli MG1655, and --

The results are in!

Fig 1: Effects of errors on assembly graph density at radius=10 for E. coli MG1655.

Basically, we see the same effect as with Fig 1 in the last post: when there are more errors in the second half of the read, the average local graph connectivity is lower. Also note that (comparing the Y axis levels in Fig 3 from the last post to Fig 1 above) E. coli isn't very repetitive at all, which we kind of knew.

So, what could be going on?

  1. E. coli isn't repetitive enough to give us a real test. But I think it does directly address Zam's concern that the polymorphisms in IS elements and other repeats would lead to inadvertent connectivity -- it appears it's not quite that simple.
  2. What we really need are metagenome-like abundances, which is to say multiple somewhat overlapping genomes with different abundances; this will then supply the necessary graph density increase in the face of random errors. I'll be testing that next.
  3. Aliens. Some explanation we haven't thought of.
  4. Our original explanation in the assembly artifacts paper: the sequencer is sticking gunk on the end.

Obviously it's going to be hard to rule out #3, but I think we lay out a pretty strong argument for #4 in the paper, at least once we can rule out #2 and previous.

--titus

Comments !

(Please check out the comments policy before commenting.)