fillet.md

Introduction

duplex_tools split_on_adapter is a tool for splitting Oxford Nanopore Sequencing reads based on knowledge of adapter sequences. Its primary use case is for splitting chimeric reads into their component sub-reads.

Installation

split_on_adapter is included when performing the installation of duplex-tools from PyPI with:

pip install duplex-tools

Usage

Run it like this (sample_type can be Native or PCR):

duplex_tools split_on_adapter <fastq_directory> <output_directory> <sample_type>

To see more options run:

duplex_tools split_on_adapter --help

For each input fastq file found in the input directory, a file with an additional _split suffix will be output into the output directory.

The new *_split.fastq will contain two new reads for each read that was split, now with suffix _1 and _2 Reads which were not split will also be added to the new file, so that *_split.fastq can be used as a match for any downstream analysis. For example the output may look like:

@<read_id> <remaining_headers>
<sequence>
+
<quality>
@<read_id>_1 <remaining_headers>
<sequence>
+
<quality>
@<read_id>_2 <remaining_headers>
<sequence>
+
<quality>

Algorithmic Details

The purpose of read_fillet is to split reads likely to be concatemers of independent reads. Such concatemers may arise from at least two mechanisms:

• chemical chimeras arising from artifacts of molecule-biological steps used in the preparation of sequencing libraries,
• informatic chimeras arising from failures of the algorithms used to process the primary sequencing data.

Regardless of the mechanism that has created chimeric reads, read_fillet attempts to split such reads into their component sub-reads using knowledge of the sequencing adapters used in library preparation. Concatemeric reads typically contain one or more matches to a sequencing adapter internal to their sequence; for example, commonly half-way through the read when library read-length is from a tightly centered distribution. The following simplified views indicate the types of error that may occur.

Here is a read, where >> represents the adapter sequence and = represent bases from the organism of interest. The reverse-complement of the sequencing adapter can typically be found at the end of a read, represented as << below:

    >>=====<<

A simple chimeric read containing two sub-reads can be represented in this notation as:

        A        B
    >>=====<<>>=====<<

On Oxford Nanopore sequencing platforms, and enzyme is used to unwind double-stranded DNA and thread a single strand through a nanopore. It is possible that after the first strand of a duplex has transited the nanopore, that the second strand is immediately captured since it is physically close to the pore. If the capture is particularly fast the sequencing device software does not detect a boundary between the signals measured for the first and second strand and subsequently produced a single conjoined read. In this case the sub-read B will be complementary to A. It is possible also that the sub-read B is distinct from A, being derived from a second DNA duplex unrelated to that which gave rise to read A. In this case, depending on the sequencing library, the sequences of A and B can be unrelated.

Read-fillet looks for matches to the sequence <<>> and splits the original basecall for the read into separate parts, for example the read:

Before:

    read_id: 0ae195a2-6993-4a0b-afa8-bb834f4739e3

            A        B
        >>=====<<>>=====<<
               ^--^

wil become:

    read_id: 0ae195a2-6993-4a0b-afa8-bb834f4739e3_1
            A
        >>=====<<

    read_id: 0ae195a2-6993-4a0b-afa8-bb834f4739e3_2
            B
        >>=====<<

after the splitting process. Note the _1 and _2 in the emitted read UUIDs.

Assessment

An additional program assess_split_on_adapter is available to provide an assessment of the veracity of the results provided by the main duplex_tools split_on_adapter program. The assessment program uses alignment of reads to a reference sequence to form knowledge of the true structure of reads (assuming the reference sequence to be correct and, for example, not contain structural variants with respect to the sequences sample).

The assessment program requires the addition dependencies: pomoxis, samtools, and seqkit. These are most easily obtained using conda (or mamba as a faster alternative):

mamba create --name duplex_env -c bioconda seqkit samtools pomoxis python3.6
conda activate duplex_env
pip install duplex-tools

The program examines supplementary alignments produced by the minimap2 aligner. The alignments are checked to determine their overlap to both the read and the reference sequence. Reads are grouped into one of the following classes:

• single_alignment_95%cov: the read is associated with a single alignment spanning >95% of the reads length,
• disjoint_with_gap: two alignments are found for the read with an unaligned, adapter-sized gap excluded from the alignments,
• disjoint_without_gap: similar to the former, without an adapter-sized gap,
• overlapping: two alignments are found which overlap each other with respect to the reference.
• read_gt_2_supplementary: the read is associated with more than two alignments.

These categories are not mutually exclusive, if a read belongs to multiple classes the priority of labelling is from top to bottom as written above.

To assess the veracity of the adapter-based duplex_tools split_on_adapter read splitting the assessment program counts reads belonging to with- and without-gap classes as follows:

disjoint_with_gap

• split_on_adapter split: true positive
• split_on_adapter unsplit: false negative

disjoint_without_gap

• split_on_adapter split: false positive
• split_on_adapter unsplit: true negative

Help

Licence and Copyright

© 2021- Oxford Nanopore Technologies Ltd.

duplex_tools is distributed under the terms of the Mozilla Public License 2.0.

Research Release

Research releases are provided as technology demonstrators to provide early access to features or stimulate Community development of tools. Support for this software will be minimal and is only provided directly by the developers. Feature requests, improvements, and discussions are welcome and can be implemented by forking and pull requests. However much as we would like to rectify every issue and piece of feedback users may have, the developers may have limited resource for support of this software. Research releases may be unstable and subject to rapid iteration by Oxford Nanopore Technologies.