Aligning Nanopore reads against a highly similar reference.

[coro1c]

Hello,

We are doing high-throughput cloning of different gene variants and want to move the QC of our cloning products to Nanopore sequences. I did a trial run with a library of ~400 variants. The variants have large constant parts but differ in two 40 bp stretches.

When I try to align the Nanopore reads from a flongle (N50: 2.14kb, avg Q-score: 18), I get large alignment errors most likely due to the reference being so similar. I tried several flags (-U, -f to limit kmers in common regions, -G to limit gap length to a few bp, ...). However, my alignment either crashes or gives low quality alignments in the variable regions (in the picture: ~350 and ~1350 bp).

Can you maybe give some advice/recommendations, how I can best tackle this alignment problem?

Many thanks in advance.

Best,
Marie

[lh3]

Are you using 400 sequences as the reference? What command line are you using exactly? Why do you think the alignment is wrong, instead of the cloning being wrong?

[coro1c]

Yes, I use 400 sequences as reference. I am using the default settings of EPI2ME wf-alignment, so I think these are the -x map-ont settings.

I checked for a some reads manually and they actually map back to another reference with some indels.
So I cannot fully rule out cloning errors but for all reads I checked (~20) I got actually a better fitting reference. I think the largest problem are actually indels due to the low Q score of 18.

Many thanks for your help.

[lh3]

Please run minimap2 independently and provide the exact command line. Also what do you mean by "better"? Is the alignment score higher? Eyeballing doesn't really count.

[veghp]

We regularly use minimap2 in our pipeline on Nanopore data for DNA cloning (plasmid construct) QC without issues. Such a "cliff" in the coverage suggests that there are two main types of DNA products in your sample, with a large common element. We always start the analysis by running Nanoplot on the raw reads: having multiple peaks on the read length histogram suggests a polyclonal sample (on top of the obvious 400 variants which should have the same length. That approach works when you have at least some full-length reads).