Demultiplexing Nanopore reads with LAST

Note: This approach has now been bundled up into a single Perl script, fastq-dental.pl which can be found here. The steps below demonstrate how to carry out similar operations manually using the command line. here. The steps below demonstrate how to carry out similar operations manually using the command line.

Here is an example for running the fastq-dental script:

./fastq-dental.pl -barcode barcode_full_PBK004.fa -mat bc.mat reads_all.fastq.gz

Example output count information, demonstrating that reads are primarily mapped to BC06 or BC07 (i.e. the adapters that were added during sample preparation):

         9 BC01
        39 BC02
         1 BC02_BC07
        16 BC03
        38 BC04
         2 BC04_BC06
         1 BC04_BC07
         1 BC04_BC09
       643 BC05
         2 BC05_BC07
    493175 BC06
       115 BC06_BC06
         2 BC06_BC07
    663161 BC07
         1 BC07_BC05
       117 BC07_BC07
         1 BC07_BC09
       106 BC08
         1 BC08_BC06
        75 BC09
        72 BC10
         1 BC10_BC07
        41 BC11
        44 BC12
         1 BC12_BC06
     13535 BCchim
    237796 BCnoadapt
       152 BCnone
        90 RB12A

Combine all input reads into a single file

pv ../called_all/*.fastq | gzip > reads_all.fastq.gz
```Note: I'm using the pipe viewer command  _pv_  to produce a progress indicator while the command is running. If this command is not available, it can be replaced with  _cat_  with no change in function (apart from not showing progess).

Prepare a FASTA file containing barcode sequences, ideally including adapters (see attached FASTA file).

Using adapters reduces both the false negative rate, and false positive as it relates to barcode sequences being accidentally mapped to input non-adapter sequences, with some increase in false positive mapping for non-library barcode sequences. I have found that the new LAST model, RY4 , together with lastal --split , substantially reduce these non-library barcode hits to similar or better levels than a mapping of the unique component of the barcode sequences alone.

[use the adapter-excluded barcode_base_96.fa if other sequences are not appropriate]

barcode_full_PBK004.fa barcode_full_RBK004.fa barcode_full_RBK110.96.fa barcode_base_96.fa

Create a shell variable containing this file name:

barcodeFile="barcode_full_PBK004.fa"

Newer versions of LAST (v1409+) include a new seeding scheme, '-uRY4' [and other related RYX schemes], which improves mapping accuracy and reduces polyA matches; low-complexity regions are also converted to lower case. This will generate seven additional files of the form .XXX:

lastdb -uRY4 -R01 ${barcodeFile} ${barcodeFile}

Prepare a substitution matrix for barcode mapping. The default substitution matrix is swayed too much by INDELs in the barcode sequences, so here's one that I've developed using last-train using super-accuracy Nanopore reads called using Guppy v5.1.15:

#last -Q 1
#last -t4.22004
#last -a 16
#last -A 15
#last -b 4
#last -B 5
#last -S 1
# score matrix (query letters = columns, reference letters = rows):
       A      C      G      T
A      6    -33    -28    -34
C    -33      6    -34    -34
G    -14    -33      6    -34
T    -35    -34    -34      5

bc.mat

I generated this matrix from the following command:

last-train -Q 1 -P 10 ${barcodeFile} reads_all.fastq.gz > bc.mat

This quality-adjusted matrix has a small penalty for opening gaps (i.e. insertions and deletions), and a smaller penalty for inserting them. Insertions and deletions are considered to be equally likely in the barcode region. It also has a moderate penalty for A/G transition variants, with other substitution penalties similar.

Use LAST in split mode, using the pre-defined substitution matrix to map the reads. In this example, it is distributed over 10 processing threads (-P 10). Here maf-convert is used to convert to a single line per match, cut retains only the barcode and read IDs, and uniq is used to make sure that multiple same barcodes per read (e.g. for reverse / complement barcodes at each end) will not produce duplicates:

lastal --split -p bc.mat -P 10 ${barcodeFile} <(pv reads_all.fastq.gz) | \
  maf-convert -n tab | cut -f 2,7 | sort | uniq | \
  gzip > barcode_assignments.txt.gz

Stringency can be altered by adjusting the query letters per random alignment setting (-D , 1e6 by default). Lowering this number will produce more matches, at the expense of more false positive matches:

lastal --split -D 1e5 -p bc.mat -P 10 ${barcodeFile} <(pv reads_all.fastq.gz) | \
  maf-convert -n tab | cut -f 2,7 | sort | uniq | \
  gzip > barcode_assignments.txt.gz

The output of this command will be a gzipped tab-separated 2-column file with barcode names in the first column, and read IDs in the second column.

For each discovered barcode, using the appropriate read category assignment file, find the corresponding read IDs, then extract those IDs out of the read FASTQ file. This uses one of my scripts, fastx-fetch.pl, to do this directly from a FASTQ file. The ' -lengths ' command-line parameter also outputs sequence lengths for each read (see next step):

fastx-fetch.pl

mkdir -p demultiplexed
fastx-fetch.pl -lengths -demultiplex barcode_assignments.txt.gz \
  -prefix 'demultiplexed/reads' <(pv reads_all.fastq.gz) > barcode_counts.txt

Note: this demultiplexing code will only by default put reads into a barcode bin if they have a single unique barcode sequence detected. Otherwise, they will be put into a 'BCchim' bin if multiple adapters are detected (i.e. a chimeric read), or a 'BCmiss' bin if no adapters are detected. If these reads should be duplicated and put in one bin per barcode, then the -chimeric option can be added to the command arguments:

mkdir -p demultiplexed
fastx-fetch.pl -lengths -demultiplex barcode_assignments.txt.gz -chimeric \
  -prefix 'demultiplexed/reads' <(pv reads_all.fastq.gz) > barcode_counts.txt

The lengths output from the demultiplexing step can be fed into another one of my scripts, length-plot.r, in order to display length-based QC plots:

length_plot.r demultiplexed/lengths_*.txt.gz
```As output, this produces a multi-page PDF file,  _Sequence_curves.pdf_ . Here are some examples of the plots that are produced:





 **1. Read Count Frequency Curve** 





<img src="https://static.yanyin.tech/literature_test/protocol_io_true/protocols.io.14egnxw4zl5d/Sequence_curves_00.png" alt="Read count frequency curve for five samples, showing a variety of different read length distributions" loading="lazy" title="Read count frequency curve for five samples, showing a variety of different read length distributions"/>



 **2. Called Bases Frequency Curve** 





  

<img src="https://static.yanyin.tech/literature_test/protocol_io_true/protocols.io.14egnxw4zl5d/Sequence_curves_01.png" alt="Called bases frequency curve for five samples, showing a variety of different read length distributions" loading="lazy" title="Called bases frequency curve for five samples, showing a variety of different read length distributions"/>



 **3. Called Bases Density Curve** 





<img src="https://static.yanyin.tech/literature_test/protocol_io_true/protocols.io.14egnxw4zl5d/Sequence_curves_02.png" alt="Sample-normalised called bases density curve for five samples, showing a variety of different read length distributions" loading="lazy" title="Sample-normalised called bases density curve for five samples, showing a variety of different read length distributions"/>

 **4. Cumulative Sequenced Bases Curve** 





<img src="https://static.yanyin.tech/literature_test/protocol_io_true/protocols.io.14egnxw4zl5d/Sequence_curves_03.png" alt="Sample-normalised cumulative sequenced base curve for five samples, showing a variety of different read length distributions" loading="lazy" title="Sample-normalised cumulative sequenced base curve for five samples, showing a variety of different read length distributions"/>



 **5. Digital Electrophoresis Plot (relative frequency)** 





<img src="https://static.yanyin.tech/literature_test/protocol_io_true/protocols.io.14egnxw4zl5d/Sequence_curves_04.png" alt="Relative digital electrophoresis plot for five samples, showing a variety of different read length distributions" loading="lazy" title="Relative digital electrophoresis plot for five samples, showing a variety of different read length distributions"/>



 **5. Digital Electrophoresis Plot (sample-normalised)** 





<img src="https://static.yanyin.tech/literature_test/protocol_io_true/protocols.io.14egnxw4zl5d/Sequence_curves_05.png" alt="Sample-normalised digital electrophoresis plot for five samples, showing a variety of different read length distributions" loading="lazy" title="Sample-normalised digital electrophoresis plot for five samples, showing a variety of different read length distributions"/>

Following on from here, cDNA reads can be oriented in preparation for stranded mapping:

• Preparing Reads for Stranded Mapping Plasmid DNA sequences can be mapped or assembled:

• Plasmid Sequence Analysis from Long Reads