NanoVar - Structural variant caller using low-depth long-read sequencing

NanoVar is a genomic structural variant (SV) caller that utilizes low-depth long-read sequencing such as Oxford Nanopore Technologies (ONT). It characterizes SVs with using only 4x depth sequencing for homozygous SVs and 8x depth for heterozygous SVs.

Basic capabilities

• Performs long-read mapping (Minimap2) and SV discovery in a single pipeline.
• Accurately characterizes SVs using long sequencing reads (High SV recall and precision in simulation datasets, overall F1 score >0.9)
• Characterizes six classes of SVs including novel-sequence insertion, deletion, inversion, tandem duplication, sequence transposition (TPO) and translocation (TRA).
• Requires 4x and 8x sequencing depth for detecting homozygous and heterozygous SVs respectively.
• Rapid computational speed (Takes <3 hours to map and analyze 12 gigabases datasets (4x) using 24 CPU threads)
• Approximates SV genotype
• Identifies full-length LINE and SINE insertions (Marked by "TE=" in the INFO column of VCF file)

Getting Started

Quick run

nanovar [Options] -t 24 -f hg38 sample.fq/sample.bam ref.fa working_dir 

Parameter	Argument	Comment
-t	num_threads	Indicate number of CPU threads to use
-f (Optional)	gap_file (Optional)	Choose built-in gap BED file or specify own file to exclude gap regions in the reference genome. Built-in gap files include: hg19, hg38 and mm10
-	sample.fq/sample.bam	Input long-read FASTA/FASTQ file or mapped BAM file
-	ref.fa	Input reference genome in FASTA format
-	working_dir	Specify working directory

See wiki for entire list of options.

Output

Output file	Comment
${sample}.nanovar.pass.vcf	Final VCF filtered output file (1-based)
${sample}.nanovar.pass.report.html	HTML report showing run summary and statistics

For more information, see wiki.

Operating system:

• Linux (x86_64 architecture, tested in Ubuntu 14.04, 16.04, 18.04)

Installation:

There are three ways to install NanoVar:

Option 1: Conda (Recommended)

# Installing from bioconda automatically installs all dependencies 
conda install -c bioconda nanovar

Option 2: PyPI (See dependencies below)

# Installing from PyPI requires own installation of dependencies, see below
pip install nanovar

Option 3: GitHub (See dependencies below)

# Installing from GitHub requires own installation of dependencies, see below
git clone https://github.com/cytham/nanovar.git 
cd nanovar 
pip install .

Installation of dependencies

• bedtools >=2.26.0
• samtools >=1.3.0
• minimap2 >=2.17

Please make sure each executable binary is in PATH.

1. bedtools

Please visit here for instructions to install.

2. samtools

Please visit here for instructions to install.

3. minimap2

Please visit here for instructions to install.

Documentation

See wiki for more information.

Versioning

See CHANGELOG

Citation

If you use NanoVar, please cite:

Tham, CY., Tirado-Magallanes, R., Goh, Y. et al. NanoVar: accurate characterization of patients’ genomic structural variants using low-depth nanopore sequencing. Genome Biol. 21, 56 (2020). https://doi.org/10.1186/s13059-020-01968-7

Authors

• Tham Cheng Yong - cytham
• Roberto Tirado Magallanes - rtmag
• Touati Benoukraf - benoukraflab

License

This project is licensed under GNU General Public License - see LICENSE.txt for details.

Simulation datasets and scripts used in the manuscript

SV simulation datasets used in the manuscript can be downloaded here. Scripts used for simulation dataset generation and tool performance comparison are available here.

Although NanoVar is provided with a universal model and threshold score, instructions required for building a custom neural-network model is available here.

Limitations

• The inaccurate basecalling of large homopolymer or low complexity DNA regions may result in the false determination of deletion SVs. We advise the use of up-to-date ONT basecallers such as Dorado to minimize this possibility.

• For BND SVs, NanoVar is unable to calculate the actual number of SV-opposing reads (normal reads) at the novel adjacency as there are two breakends from distant locations. It is not clear whether the novel adjacency is derived from both or either breakends in cases of balanced and unbalanced variants, and therefore it is not possible to know which breakend location(s) to consider for counting normal reads. Currently, NanoVar approximates the normal read count by the minimum count from either breakend location. Although this helps in capturing unbalanced BNDs, it might lead to some false positives.