ColabFold¶

Easy to use protein structure and complex prediction using AlphaFold2 and Alphafold2-multimer.

Local ColabFold is made available on the cluster as a shared Apptainer container image. This should be run on a GPU Compute partition.

You can use the apptainer/colabfold module which will add convenient aliases for: colabfold_batch, colabfold_search, colabfold_split_msas, which will run within the container; The alias will also bind-mount the AlphaFold2 weights cache path (/opt/colabfold/alpha2_weights_cache on the nodes) into the container on /cache.

See the LocalColabFold documentation for usage information. Example:

Terminal

module load apptainer/colabfold
# The following is required to use aliases in a non-interactive/SLURM batch script:
shopt -s expand_aliases
colabfold_batch ./input.fasta ./out/

An example Slurm script to run ColabFold on the cluster is provided below:

Terminal

#!/bin/bash

#SBATCH --job-name=colabfold    # Job name
#SBATCH --partition=aoraki_gpu  # Partition (queue) name
#SBATCH --nodes=1               # Number of nodes
#SBATCH --ntasks-per-node=1     # Number of tasks (1 task per node)
#SBATCH --cpus-per-task=12      # Number of CPU cores per task
#SBATCH --gres=gpu:1            # Number of GPUs required
#SBATCH --mem=96G               # Job memory request
#SBATCH --time=10:00:00         # Time limit hrs:min:sec
#SBATCH --mail-user=USERNAME@otago.ac.nz
#SBATCH --output=colabfold%j.log # Standard output log

# Set variables
base_name="$1"
output_fasta="${base_name}_getorf.output.fa"

# Load the apptainer/colabfold module (assuming it's in your PATH)
module load apptainer/colabfold

shopt -s expand_aliases  # Enable alias expansion

# Check loaded modules (for debugging)
module list

# Ensure PATH includes module binaries (for debugging)
echo "Current PATH: $PATH"

# Run colabfold_batch command using alias
colabfold_batch "./$output_fasta" ./out/