Demultiplexing SRA data

Question

[Adrian Pelin]

Hello,

I am looking at this experiment

Looks like the authors uploaded a multiplexed dataset with no info on barcoding. Any way to guess the barcodes? Just by running fastqc you can sorta guess what some barcodes may be, but is there a better way?

I downloaded the sra file and used --split-3, which gave me 2 files for PE.

Here is what the file sorta looks like:

@SRR2046220.999996 HISEQ:108:C24D5ACXX:4:1101:5712:37823 length=102
AAGGATGCAATTCCTGGTGGTGCCATGGAGGTAAAGTCATAGTATTTTTATGATTTATATTTACATATTTTTACACTTCATAGTCATTTTTATAAAACTTTN
+SRR2046220.999996 HISEQ:108:C24D5ACXX:4:1101:5712:37823 length=102
CCCFFFFFHHHHHJJJIFGGFHIIHIIJAHIEGHJHFGGFIIBGIIJJJJJIJJIJJJJIJJJJJJIJIIJJJIIHHIIJIHHHFHHEFFFFFFEDDECEE#
@SRR2046220.999997 HISEQ:108:C24D5ACXX:4:1101:5536:37823 length=102
CAGGACGAAAATGAAGGTTTGGTTTTAACATTTGATCTGAGTTTATAGTATAGAAAGAGATCTATATTGACTCAGCTTTGCATATAAATCATACATTCTAGN
+SRR2046220.999997 HISEQ:108:C24D5ACXX:4:1101:5536:37823 length=102
######################################################################################################
@SRR2046220.999998 HISEQ:108:C24D5ACXX:4:1101:5653:37824 length=102
TAACTCTCTATTCACGAAAATCTGATCAATTGGATGACGGCTCGAAGAGCTTGATTCTACCAGATAGTACAGTTACATCAGGATGAAGTGCAGAAACGCTTN
+SRR2046220.999998 HISEQ:108:C24D5ACXX:4:1101:5653:37824 length=102
0;8@##################################################################################################
@SRR2046220.999999 HISEQ:108:C24D5ACXX:4:1101:5739:37825 length=102
CTCAATTCAATTCGGAGCTTCGTCCCCTACAGGACCTCACCCTTCGATCAAACTAAATTATTATTCTTTTTCCAATATTACAATATCAACAATATGTACGTN
+SRR2046220.999999 HISEQ:108:C24D5ACXX:4:1101:5739:37825 length=102
1++44=BDF>FFFEEG1CFCGIDGHFH@G>FGEHDGGIIIIID;;FFHEG8@FG@AGHH;CAEFFEHHHEEEDDE>CCEDCA5>>CDEC?@?CCCC>@CB<#
@SRR2046220.1000000 HISEQ:108:C24D5ACXX:4:1101:5569:37827 length=102
TCTCTTACAATTCCAAAAGATATAGATAAGGCAATTTATTGGTATGAAGAATCTGCTAAACAAGGAAATCAAGGTGCACAAAATAGTTTAGAAGGACTTCAN
+SRR2046220.1000000 HISEQ:108:C24D5ACXX:4:1101:5569:37827 length=102
+++22?@A+?CCCCBBBCBCABBBBBCBBBBCCBBBBBBBBBABABBBBBBBBBBBBBBBBBBBBBABBBBABB>=ABAAAA<>>@?@@B>@@@@==;???#

[GenoMax]

Based on the SRA entry this is ddRAD-Seq data. ENA also has just paired end fastq files. This must be EcoRI-MspI digested fragments.

[Arindam Ghosh]

How to detect if a certain SRA RNA-seq fastq file has been demultiplexed or not?

[GenoMax]

Check the index sequences in fastq headers. If there are more than one the file is likely not demultiplexed.

Answer 1

[Istvan Albert]

I have a semi empirical "method" cut the first N bases from the first 100,000K sequences then see how many unique you get. Something like (untested just typed it in)

cat data.fq | head -100000 |\
    bioawk -c fastx ' { print substr($seq,1,8) }' | sort |\
    uniq -c | sort -rn

Then keep raising the 8 to 9, 10 etc as long as you read the indexed adapter you'll have clear groups with approximately the same number of sequences.

[Adrian Pelin]

Very very cool idea, love it! I tried on 1 million

So, at 8 got:

1490 TCAATATC
1376 TAAAAAAA
1351 GGCATATC
1144 TGATCGCC
1063 GGACTCAC
1055 GGATATAC
1026 GGCAAGGC
1006 GGCCATCC

Low complexity like TAAAAAA is probably a false one.

At 9:

1465 TCAATATCAA
1347 GGCATATCAA
1130 TGATCGCCAA
1039 GGATATACAA
 996 GGCAAGGCAA
 984 GGCCATCCAA
 965 GGACTCACAA
 907 GAACTTGCAA

Interesting that all end in AA.

At 15:

457 TAANNNNNNNNNNNN
187 TCAATATCAATTCAA
182 TCAATATCAATTCTT
166 GGCATATCAATTCCT
157 TCAATATCAATTCAT
149 TGATCGCCAATTCAA
143 GGCATATCAATTCAA
142 GGCAAGGCAATTCAA
134 GGCCATCCAATTCAA
133 GGATATACAATTCAA
128 GGATATACAATTCTT

Is everything before AA the index sequence?

[Istvan Albert]

normally there should be a big drop in the groupings, but make sure to use it on the first file, only that will have the index, the other pair won't

[Adrian Pelin]

Yeah I am using it on _1 from sra. Will try on _2. A big drop in grouping would make sense, but I don't think I see it. It may be that the file was demultiplexed, barcodes removed, and then all groups merged back together? That hardly makes sense but maybe possible.

[Istvan Albert]

yeah, your data does not seem to indicate the presence of common sequences at the start.