Differences in PIPseeker official results?

[jamesnemesh]

I noticed in the PIPSeeker documentation there's now an ability to extract converted fastq reads, in the same sort of way your program does.

It would be optimal for my peace of mind if the two programs gave the same outputs.

You can test this by converting your example data to have the suffix .fastq.gz, for example with a soft link:
example_R1.fastq.gz -> example_R1.fq.gz

Then run PipSeeker:
pipseeker barcode --fastq data/example_v3/. --output-path PIPseeker --chemistry v3

An R1 read from PIPSeeker:

@A01831:50:HCLHTDRX3:1:2101:16495:1000 1:N:0:TAAGGCGA
AGTACCGGCACGAAGTGGCGTCCGAACG
+
~~~~~~~~~~~~~~~~FFFFFFFFF:FF

An R1 read from your software (using your tutorial command to generate the data):

@A01831:50:HCLHTDRX3:1:2101:16495:1000 1:N:0:TAAGGCGA
AGAGGTGCACAGCAACAAAGACAAGTAGGGCGTCCGAACG
+
FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF:FF```

You both agree on the UMI sequence, but your barcoded sequence is significantly longer than what PIPseeker emits.

AGTACCGGCACGAAGT GGCGTCCGAACG
AGAGGTGCACAGCAACAAAGACAAGTAG GGCGTCCGAACG

I have no idea what black voodoo magic they are performing to get to a 16bp barcode.

[noamteyssier]

Hm that's interesting - I don't know how they would be able to get a 16bp barcode either considering the barcode lengths themselves are 8+6+6+8=28.

I guess the major question to answer is whether the number of unique cells changes between the two. If they don't maybe they just create a unique 16mer to be more compatible with existing software (i.e. you don't need to adjust the default technology flag for kallisto or salmon).

[zaki-bds]

Hi all!

I was plagued with a similar problem, as to how we map from 28bp barcode generated by pipspeak, to the 16bp barcode generated by pipseeker. This was the response I received from their bioinformatics team:

Hi Zaki,

Within each tier, the barcode index is translated into a 4-base sequence:

0 ==> AAAA, 1 ==> AAAC, 2 ==> AAAG, 3 ==> AAAT, 4 ==> AACA, 5 ==> AACC, ..., 94 ==> CCTG, 95 ==> CCTT

The sequences from each of the 4 tiers are joined into a 16-base sequence.

The barcode index that they are referring to is the same that is provided in the tsv files with pipspeak. This approach solved the problem for me.