Scientific Software

Scientific applications available on the cluster.

AlphaFold

AlphaFold is an AI program developed by DeepMind that predicts protein structures from amino acid sequences.

Databases

The necessary databases are pre-mounted on GPU nodes at /alphafold_storage/alphafold_db — no download needed.

Required Parameters

Parameter Description
-d <data_dir> Path to the supporting data directory
-o <output_dir> Path to store results
-f <fasta_paths> Path to FASTA file(s). Multiple sequences in one file = multimer. Multiple files comma-separated = fold sequentially
-t <max_template_date> Maximum template release date (YYYY-MM-DD)

Optional Parameters

Parameter Default Description
-g true Enable NVIDIA GPU runtime
-r true Run final relaxation step
-e true Run relax on GPU
-n all cores OpenMM threads
-a 0 CUDA_VISIBLE_DEVICES — comma-separated GPU list
-m monomer Model preset: monomer, monomer_casp14, monomer_ptm, multimer
-c full_dbs MSA database preset: reduced_dbs or full_dbs
-p false Use precomputed MSAs from disk
-l 5 Predictions per model (multimer only)
-b false Benchmark mode — excludes compilation time

Example Job Script

#!/bin/bash
#SBATCH --job-name=AlphaFold-Multimer
#SBATCH --partition=gpu2
#SBATCH --nodes=1
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=4
#SBATCH --mem=32G
#SBATCH --gres=gpu:1
#SBATCH --output=alphafold_%j.out
#SBATCH --error=alphafold_%j.err

module load alphafold/alphafold_non_docker_2.3.1

bash $ALPHAFOLD_SCRIPT_PATH/run_alphafold.sh \
  -d $ALPHAFOLD_DB_PATH \
  -o ~/output_dir \
  -f $ALPHAFOLD_SCRIPT_PATH/examples/query.fasta \
  -t $(date +%Y-%m-%d)

Memory Guidelines

Additional Resources

AlphaFold3

AlphaFold3 runs via Singularity/Apptainer container on GPU nodes with the af3 GRES constraint.

Loading the Module

module load alphafold3

This automatically loads Apptainer and sets the following environment variables:

Variable Description
$AF3_CONTAINER Path to the Singularity container
$AF3_MODELS Path to model parameters
$AF3_DB Path to the database
$AF3_SRC Path to AlphaFold3 source directory

Bind Mounts

Singularity requires bind mounts to expose host directories inside the container:

--bind /path/on/host:/path/inside/container

Your input folder can be bound to any path inside the container — /root/af_input is not required, it's just an example.

Running AlphaFold3

singularity exec --nv \
    --bind $AF3_SRC:/root/custom_folder \
    --bind /tmp:/root/af_output \
    --bind $AF3_MODELS:/root/models \
    --bind $AF3_DB:/root/public_databases \
    --bind /home/user/alphafold_inputs:/root/custom_folder \
    $AF3_CONTAINER \
    python /root/custom_folder/run_alphafold.py \
        --json_path=/root/custom_folder/fold_input.json \
        --model_dir=/root/models \
        --db_dir=/root/public_databases \
        --output_dir=/root/af_output

Replace /home/user/alphafold_inputs with the actual path to your input folder.

For input file format, see the AlphaFold3 Input File Guide.

Listing All Available Flags

singularity exec --nv \
    --bind $AF3_SRC:/root/custom_folder \
    --bind /tmp:/root/af_output \
    --bind $AF3_MODELS:/root/models \
    --bind $AF3_DB:/root/public_databases \
    $AF3_CONTAINER \
    python /root/custom_folder/run_alphafold.py --helpfull

Example Job Script

#!/bin/bash
#SBATCH --job-name=alphafold3
#SBATCH --partition=gpu-general
#SBATCH --gres=gpu:1,af3
#SBATCH --cpus-per-task=8
#SBATCH --mem=64G
#SBATCH --time=1-00:00:00
#SBATCH --output=alphafold3_%j.out
#SBATCH --error=alphafold3_%j.err

module load alphafold3

singularity exec --nv \
    --bind $AF3_SRC:/root/custom_folder \
    --bind /tmp:/root/af_output \
    --bind $AF3_MODELS:/root/models \
    --bind $AF3_DB:/root/public_databases \
    --bind /home/user/alphafold_inputs:/root/custom_folder \
    $AF3_CONTAINER \
    python /root/custom_folder/run_alphafold.py \
        --json_path=/root/custom_folder/fold_input.json \
        --model_dir=/root/models \
        --db_dir=/root/public_databases \
        --output_dir=/root/af_output

Unloading

module unload alphafold3

This also unloads the Apptainer module.

Troubleshooting

RELION

RELION is a cryo-EM structure determination package. On the TAU HPC cluster it runs on the dedicated gpu-relion partition with X11 forwarding.

Requirements

Starting a RELION Session

Start an interactive job on the GPU RELION partition with X11:

srun --ntasks=1 -p gpu-relion-pool -A gpu-relion-users_v2 --qos=owner --x11 --pty bash

Load the RELION module:

module load relion/relion-4.0.1

Launch RELION:

relion

Notes