ONTAP

ONT Amplicon sequencing Pipeline

A Nextflow pipeline for basecalling, read mapping, QC, variant calling and analysis of nanopore multiplex amplicon data.

For tool versions please see the processes within the 'modules' directory.

Installation

1. Install Nextflow

2. Install Docker

3. Download the appropriate Dorado installer from the repo. The path to the executable will be <path to downloaded folder>/bin/dorado

4. (Optional) Download the appropriate Dorado model from the repo

   # Download all models
   dorado download --model all
   # Download particular model
   dorado download --model <model>
   

   If a pre-downloaded model path is not provided to the pipeline, the model specified by the --basecall_model parameter will be downloaded on the fly.

5. Download the appropriate Clair3 model from the Rerio repo (you will need Python3)

   First clone the repo:

   git clone https://github.com/nanoporetech/rerio
   

   This contains scripts to download the model(s) to clair3_models/<config>

   #  Download all models
   python3 download_model.py --clair3
   #  Download particular model
   python3 download_model.py --clair3 clair3_models/<config>_model
   

   Each downloaded model can be found in the repo directory under clair3_models/<config>

6. Clone the repository with required submodules

   git clone --recurse-submodules https://github.com/sanger-pathogens/ONTAP.git
   

Usage

nextflow run ONTAP/main.nf \
--raw_read_dir <directory containing FAST5/POD5 files> \
--reference <reference fasta> \
--primers <fasta containing primers> \
--target_regions_bed <BED file containing target regions> \
--additional_metadata <CSV mapping sample IDs to barcodes> \
--dorado_local_path <absolute path to Dorado executable> \
--clair3_model <path to Clair3 model> \
-profile docker

The examples folder contains some example files.

Instead of -profile docker, you can run the pipeline with -profile laptop. As well as enabling docker, the laptop profile allows the pipeline to be used offline by providing a local copy of a configuration file that is otherwise downloaded.

Should you need to run the pipeline offline, it is best to make use of pre-populated dependency caches. These can be created with any of the supported profiles (e.g. -profile docker) by running the pipeline once to completion. You will also need to provide a --basecall_model_path (see step 4 above)- the laptop profile includes a default local path for this, as well as --clair3_model and --dorado_local_path.

You can override the default paths using the command line parameters directly when invoking nextflow or by supplying an additional config file in which these parameters are set, using the -c my_custom.config nextflow option.

Demo

To run a short demo please follow the instructions found in the demo document: Here

Other parameters:

Basecalling

• --basecall = "true"
• --basecall_model = "[email protected]"
• --basecall_model_path = ""
• --trim_adapters = "all"
• --barcode_kit_name = ["SQK-NBD114-24"] (currently this can only be edited via the config file)
• --read_format = "fastq"

Saving output files

• --keep_sorted_bam = true
• --save_fastqs = true
• --save_trimmed = true
• --save_too_short = true
• --save_too_long = true

QC

• --qc_reads = true
• --min_qscore = 9
• --cutadapt_args = "-e 0.15 --no-indels --overlap 18"
• --lower_read_length_cutoff = 450
• --upper_read_length_cutoff = 800
• --coverage_reporting_thresholds = "1,2,8,10,25,30,40,50,100"
• --coverage_filtering_threshold = "25"
• --multiqc_config = ""

Variant calling

• --clair3_min_coverage = "5"
• --masking_quality = "15"

Consensus curation

• --min_ref_gt_qual = 1
• --min_alt_gt_qual = 1

Tree building

• --remove_recombination = false
• --raxml_base_model = 'GTR+G4'
• --raxml_threads = 2

Running on Sanger farm

Load nextflow and singularity modules:

module load nextflow ISG/singularity

Follow steps 5 and 6 above to download a Clair3 model and clone the repo.

Usage is slightly different (you use the standard profile and don't need --dorado_local_path):

nextflow run ONTAP/main.nf \
--raw_read_dir <directory containing FAST5/POD5 files> \
--reference <reference fasta> \
--primers <fasta containing primers> \
--target_regions_bed <BED file containing target regions> \
--additional_metadata <CSV mapping sample IDs to barcodes> \
--clair3_model <path to Clair3 model> \
-profile standard

The standard profile is intended to allow the pipeline to run (with internet access) on the Sanger HPC (farm). It ensures the pipeline can run with the LSF job scheduler and uses singularity images for dependencies management, as well as the latest versions of the pipeline base configuration (from PaM Info common config file) and Dorado models.

It's best to run the pipeline as a job in the oversubscribed queue i.e. preface the command with this:

bsub -o output.o -e error.e -q oversubscribed -R "select[mem>4000] rusage[mem=4000]" -M4000

Once your job has finished and you're happy with the output, clean up any intermediate files. To do this (assuming no other pipelines are running from the current working directory), run:

rm -rf work .nextflow*

OS requirements

The pipeline has been tested on multiple operating systems

Linux: Ubuntu 18.04 + 22.04 (Singularity and Docker profiles) Mac OSX (Laptop profile)

Hardware requirements

This pipeline requires only a standard modern computer with enough RAM/CPU power to support running the standard tools. There are checks within the pipeline to determine max resources avaliable to ensure sensible resource requests.

The most taxing sections of the pipeline are:

• FastQC
• Minimap2
• RAXML-NG
• Clair3

The pipeline will run without GPU support; however, access to GPU hardware (one that is supported by Dorado; see here) will result in a much faster runtime.

Approximate runtime

When executed in laptop mode, a full analysis run including basecalling is typically expected to complete in approximately 10–12 hours. However, this estimate is highly sensitive to sequencing depth; larger input datasets can significantly extend basecalling time. To optimize runtime, we recommend adjusting Dorado basecalling accuracy parameters in accordance with your performance requirements (fast, hac, sup)

In HPC environments utilizing Singularity containers with GPU acceleration, runtimes are substantially reduced. Under optimal conditions, end-to-end processing has been observed to complete in as little as 40 minutes, and up to 1.5 hours for typical runs. This is however depending on input size and GPU availability.

Note On first-time runs, where none of the required tools are cached, the initial setup (including downloading dpendency software and models) introduces an additional overhead of approximately 10 minutes.

Dependencies

Tool	Version	Container
bcftools	1.20	quay.io/biocontainers/bcftools1.20--h8b25389_0
bedtools	2.31.1	quay.io/biocontainers/bedtools:2.31.1--hf5e1c6e_1
clair3	v1.0.9	hkubal/clair3:v1.0.9
pysam	0.0.2	quay.io/sangerpathogens/pysam:0.0.2
pandas	2.2.1	quay.io/sangerpathogens/pandas:2.2.1
python_graphics	1.0.0	quay.io/sangerpathogens/python_graphics:1.0.0
cutadapt	4.7	quay.io/biocontainers/cutadapt:4.7--py310h4b81fae_1
cuda_dorado	0.7.1	quay.io/sangerpathogens/cuda_dorado:0.7.1 (SEE NOTE)
fastqc	0.12.1	quay.io/biocontainers/fastqc:0.12.1--hdfd78af_0
ubuntu	20.04	ubuntu:20.04
minimap2	2.26	quay.io/biocontainers/minimap2:2.26--he4a0461_2
multiqc	1.22.2	quay.io/biocontainers/multiqc:1.22.2--pyhdfd78af_0
pod5	0.3.6	quay.io/sangerpathogens/pod5:0.3.6
pycoqc	2.5.2	quay.io/biocontainers/pycoqc:2.5.2--py_0
samtools	1.19.2	quay.io/biocontainers/samtools:1.19.2--h50ea8bc_1
seqtk	1.4	quay.io/biocontainers/seqtk:1.4--he4a0461_2

NOTE however we suggest you install your own version of dorado to match your OS

Support

Please contact PaM Informatics for support through our helpdesk portal or for external users please reach out by email: [email protected]