ONT MinION mapper benchmarking

[dovah]

Hi all,

I have RNA sequencing data (D. melanogaster) from r7 and r9 flowcells. What would be your suggestion for a mapper? I'm mostly interested in benchmarking isoform detection compared to illumina HiSeq2500/PacBio RSII. I'd like to do a benchmarking of the "established" tools (since I realized this knowledge is missing), but you can suggest new ones.

Thanks in advance for suggestions.

[gringer]

bwa in nanopore mode (bwa mem -x ont2d) should be reasonable for mapping. I've had the best results from LAST (particularly for counting subsections of chimeric 2D reads), but it needs a lot of tweaking to get good results.

[dovah]

Thanks for reply, @gringer. However, may I ask you which criteria you use to say that you got best result from a given mapper? aka, what are the criteria? I can probably imagine you take into account the number of matches/mismatches per read length...but what else? I'd like to build my own benchmark. Thanks

[gringer]

Eyeballing the results. I guess my criteria for LAST tweaking is something like "length of match at an e-value threshold that looks like it would exclude noisy matches, typically >1000". When comparing different methods, it's the proportion of reads that match to the target reference genome.

[ymc]

Thanks for reply, @gringer. However, may I ask you which criteria you use to say that you got best result from a given mapper? aka, what are the criteria? I can probably imagine you take into account the number of matches/mismatches per read length...but what else? I'd like to build my own benchmark. Thanks

I think you can download Loman et al's E coli K12 MG1655 data. Supposedly all the reads should come from this genome. So you align the called fastq to the reference and see how well different mapper aligns.

[gringer]

Most people testing MinION mapping statistics at the moment are only concentrating on the reads that map, and ignoring comparisons with reads that don't map. You can get really good mapping statistics if you're only looking at the best 10% of MinION reads, for example.