Question about QC filtering

[LeeBergstrand]

For rule nanopore_qc_filter:

reformat.sh in={input} out={output} minlength={params.minlength} minavgquality={params.minavgquality} \
interleaved=f qin=33 threads={threads} -Xmx{resources.mem}G pigz=t unpigz=t > {log} 2>&1

Why are we setting minavgquality versus parameters such as qtrim and trimq?

minavgquality=0         (maq) Reads with average quality (after trimming) below this will be discarded.
qtrim=f                 Trim read ends to remove bases with quality below trimq.
                        Values: t (trim both ends), f (neither end), r (right end only), l (left end only), w (sliding window).
trimq=6                 Regions with average quality BELOW this will be trimmed.  Can be a floating-point number like 7.3.

@jmtsuji Would it make sense to qtrim the reads and then filter by length rather than dropping reads based on minavgquality? Or are nanopore error profiles not conducive to end trimming?

[LeeBergstrand]

Looks like the quality scores go down over the read length. Though some of the reads go back up.

[LeeBergstrand]

@jmtsuji I ran stats on the assemblies of 32 genomes from one of our internal datasets. All these samples used older flowcell and base caller versions, so I had to trim them using Q score 8 (I tried many Q score cut-offs, and this had the best assemblies)and assemble them using flay nano-raw mode. Here are the results:

There was not much difference between dropping the reads with a Q score less than X and end trimming the reads to a Q score of X. I assumed that you would get slightly more data into the assembly if you were end trimming but it does not appear to make much of a difference. @jmtsuji Thoughts?

[LeeBergstrand]

From Mike Lynch:

End trimming makes sense in environments like Illumina and 454/pyrosequencing where enzyme decay puts Q-score recovery in question. It was my understanding that nanopore (and definitely PacBio) don’t have the same issue. That quality decay can easily recover (in some ways the Q-score of a set of nucleotide calls is at least partially independent of previous Q-scores - this obviously doesn’t hold if the Q-score decay is due to template issues).

So, keep average score trimming, which should be more representative of nanopore error profiles.

I feel we can close this.