False positives in filter_pairs when run on linearized plasmid about duplex-tools HOT 4 CLOSED

Hi @eyesmo,
It's going to be difficult to decide whether a read is a follow-on in most cases where you're sequencing only a single read. If ther are at least a few different sequences in the sample, then these options can be used to adjust the stringency of the match:

bases_to_align:
https://github.com/nanoporetech/duplex-tools/blob/master/duplex_tools/filter_pairs.py#L39
align_threshold:
https://github.com/nanoporetech/duplex-tools/blob/master/duplex_tools/filter_pairs.py#L40

You're right that one option would be to use UMIs or something similar to identify actual pairs in these cases.

I would suggest to instead spike in some amount of other sequences in these cases to make it easier to identify correct follow-ons. Lambda is typically pretty good as a spike-in for example.

Technically, filter_pairs is quite quick-and-dirty and checks precisely whether a read is rc of the other. The core alignment method happens here, note that seq2 is already reverse-complemented earlier, which is why this works:

https://github.com/nanoporetech/duplex-tools/blob/master/duplex_tools/filter_pairs.py#L220

from duplex-tools.

Better yet, do the sequencing adapters have any UMIs/stretches of random nucleotides embedded within them?

from duplex-tools.

I think initially used filter_pairs incorrectly. I specified a directory I intended as the output for duplex basecalling, instead of the directory where all the simplex fastq reads were stored.

Running a second time with the simplex fastq reads directory specified, it now says roughly 30% of my reads are good pairs. Still higher than I'd expect for Kit 12, but lower than 90%.

from duplex-tools.

Hi, @eyesmo, just wanted to check if you had any luck trying to increase the bases_to_align? It should help fairly well with making sure you have a complement being rc of the template.

from duplex-tools.