Odd fits at learn errors step

[gdunshea]

error-fit-example.pdf
Hi,

I have a similar issue that others have come across relating to random orientations of sequences due to our experimental design in the lab and I think it is impacting error learning.

I tried to fix the orientation problem with "Decipher::OrientNucleotides()", however when using "Decipher::readDNAStringset()" with fastq files, the quality scores are ignored and re-writing the files to fastq after "Decipher::OrientNucleotides()" ends up with nonsense quality scores. I saw that in post #434 you mentioned running through the dada2 pipeline, then checking for reverse compliments by hand, so thanks for that.

I am actually a bit concerned with how my error plots look though. The estimated error rates seem to experience a pretty severe dog-leg up at higher quality scores, in some instances dipping below the line of expected error rates - and the estimated fit in this area is generally quite poor.

Have you seen this before and do you have any suggestion for parameters I could tweak to address this?

Thanks for your time and your excellent package!

Edit: Added example figure of estimated error rates

[benjjneb]

Do you know if this data has binned quality scores? And what machine was used to generate the data (e.g. MiSeq, NovaSeq...)?

[gdunshea]

Thanks your for quick reply :)

It's a NextSeq paired-end 2x 150 run. But yes, now that you say, looking at the heatmap on the plot I posted above, it does appear that the quality scores are binned. Below is an example of the first sequence in one of the files:

@NB501850:81:HKM2LBGX5:1:11101:19604:1276 2:N:0:GTGGCC
CGTCGCTCCTACCGATTGAGTGATCCGGTGAATAATTCGGACTGCAGCAATGTTTGGATCCCGAACGTTGCAGCGGAAAGTTTAGTGAACCTTATCACTTAGAGGAAGGAGAAGTCGTAACA
+
EEEEEEEEEEEEEEEAE/EEEE/EEEEEEEAAEEEEEEEEEAEEEAEE<EEEEEEEE<EEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEAEEEAEEEEAEEAEE

[benjjneb]

See suggestions here and in the following posts on how to deal with error-rate fitting when binned quality scores are present: #938 (comment)

[gdunshea]

That's great, thanks very much Ben