Dorado

Dorado is a high-performance, easy-to-use, open source basecaller for Oxford Nanopore reads. It needs to be run on a partition or node with GPU compute/CUDA support, and is heavily optimised for NVIDIA A100 and H100 GPUs.

On the cluster, use either the aoraki_gpu_A100 or aoraki_gpu_H100 partition to ensure access to suitable GPUs.

Dorado is made available on the cluster as a shared Apptainer container image.

You can use the apptainer/dorado module to add a convenient alias to running dorado within the container:

module avail dorado
module load apptainer/dorado/0.7.1
# The following is required to use aliases in a non-interactive/SLURM batch script:
shopt -s expand_aliases
dorado ....

Alternatively, you can run Dorado directly with Apptainer; i.e. to run binaries within the container, prefix any command with apptainer -s run --nv <$APPTAINER_IMG/apptainer_image.sif>.

Make sure to specify --nv to enable NVIDIA GPU support. For example:

apptainer -s run --nv $APPTAINER_IMG/dorado-<version>.sif dorado basecaller /models/dna_r10.4.1_e8.2_400bps_hac@v4.1.0 ~/pod5s/

Also note that -s / --silent is required here to suppress the verbose container output that may otherwise contaminate standard output.

• To list the available basecaller models, run

$APPTAINER_IMG/dorado-<version>.sif ls -1 /models

As with all Apptainer containers, take care to qualify the files and paths in the context of the container image. Models are located within the container in /models/, and any other files and data stored outside the container need to be in a folder that is bound by Apptainer into the container (either by default, such as your $HOME, /scratch, or /projects), or by explicitly specifying a bind mount with Apptainer’s --bind option.

Please refer to the Dorado GitHub page for more information regarding running Dorado: https://github.com/nanoporetech/dorado

Example SLURM batch script

Below is an example SLURM batch script for running Dorado on either the aoraki_gpu_H100 or aoraki_gpu_A100 partition of the cluster. This example uses the Apptainer module with the dorado alias:

#!/bin/bash
#SBATCH --job-name=dorado_basecall
#SBATCH --partition=aoraki_gpu_A100
# Alternatively, use --partition=aoraki_gpu_H100
#SBATCH --gres=gpu:1
#SBATCH --cpus-per-task=8
#SBATCH --mem=32G
#SBATCH --time=02:00:00
#SBATCH --output=logs/dorado_%j.out
#SBATCH --error=logs/dorado_%j.err

# Load the Apptainer/Dorado module
module load apptainer/dorado/0.7.1

# Enable aliases in non-interactive SLURM shell
shopt -s expand_aliases

echo "CUDA_VISIBLE_DEVICES=$CUDA_VISIBLE_DEVICES"
nvidia-smi

# Run Dorado with the module-provided alias
dorado basecaller \
  /models/dna_r10.4.1_e8.2_400bps_hac@v4.1.0 \
  ~/pod5s/ \
  --emit-mapping