RIS-Provided Model Parameters and Input Data

A default, University-licensed copy of the model parameters and other input data is available through RIS. In order to access this data, users need to accept the terms of use. (Weights terms of use, Weights prohibited use, Output terms of use) To do this, users will need to submit a request through the RIS Service Desk acknowledging the terms of service. Users will then be granted access to the model parameters and input data. This data is located on Storage2: /storage2/fs1/RIS-Services/Active/AlphaFold3

Running AlphaFold3

The cluster-compatible Docker image is designed with default settings for the location of the model and input data. These defaults can be overridden by the user at runtime, but the input and model data locations must be mounted by the container and available to the application at runtime. The only required parameter for the application is the location for its output, which also must be mounted into the container so it is available after the job completes. The instructions that follow provide detail on running a cluster job based on the default settings and an example output directory.

General Information

Environment variables
1. ALPHAFOLD_BASE_DIR=/app/alphafold
2. RIS_SERVICES=/storage2/fs1/RIS-Services/Active/AlphaFold3
3. OUTPUT_DIRECTORY=/storage2/fs1/allocation_name/Active/alphafold3/output
  - Create /storage2/fs1/allocation_name/Active/alphafold3/outputif it doesn’t already exist
4. LSF_DOCKER_VOLUMES="$RIS_SERVICES:$RIS_SERVICES $OUTPUT_DIRECTORY:$OUTPUT_DIRECTORY $HOME:$HOME"
5. PATH="/usr/local/cuda/bin/:/opt/conda/bin:/app/alphafold:$PATH"

*allocation_name is the name of the user’s Storage allocation. Please DO NOT FORGET to replace the attributes in the above examples with appropriate values carefully.

With the environment established, the following is a simple bsub command to run the AlphaFold3 job:
```
bsub -q general-interactive -G compute-group -R "gpuhost rusage[mem=64GB]" -gpu 'num=1' -Is -a "docker(ghcr.io/washu-it-ris/alphafold3:2)" /bin/bash
```
Note that other command line options, including requests for memory and specific GPU resources are available. The example above specifically includes a request for 64 gigabytes of memory because previous test runs of the application failed to complete with lower memory limits.
At the command prompt inside the running job container, the Python virtual environment for the application needs to be activated:
```
. /alphafold3_venv/bin/activate
```
At this point the AlphaFold command is ready to run. If the RIS Service storage space for input data described above is mounted into the container along with a destination for the output data, then the defaults can be used. In this case the only required option is for the output destination:
```
python3 /app/alphafold/run_alphafold.py --output_dir=/storage2/fs1/allocation_name/Active
```

Quickstart Bsub Script

The settings and commands documented above can be combined into a quickstart script for running cluster jobs. The script below needs some small modifications to the storage locations for input and output. Additional modifications can be made to override the defaults input data locations as needed.

#!/bin/bash

export ALPHAFOLD_BASE_DIR=/app/alphafold
export RIS_SERVICES=/storage2/fs1/RIS-Services/Active/AlphaFold3
export OUTPUT_DIRECTORY=/storage2/fs1/allocation_name/Active/alphafold3/output
export LSF_DOCKER_VOLUMES="$RIS_SERVICES:$RIS_SERVICES $OUTPUT_DIRECTORY:$OUTPUT_DIRECTORY $HOME:$HOME"
export PATH="/usr/local/cuda/bin/:/opt/conda/bin:/app/alphafold:$PATH"

bsub -q general-interactive -G compute-group -R "gpuhost rusage[mem=64GB]" -gpu 'num=1' -Is -a "docker(ghcr.io/washu-it-ris/alphafold3:2)" /bin/bash

Interactive Jobs :

export PATH="/usr/local/cuda/bin/:/opt/conda/bin:/app/alphafold:$PATH" 
export ALPHAFOLD_BASE_DIR=/app/alphafold 
export RIS_SERVICES=/storage2/fs1/RIS-Services/Active/AlphaFold3 
export OUTPUT_DIRECTORY=/storage1/fs1/allocation_name/Active/alphafold3/output 
srun --pty -A compute2-group -p general-interactive --gpus=1 --container-image='ghcr.io#washu-it-ris/alphafold3:2' --container-mounts=/storage2/fs1/RIS-Services/Active/AlphaFold3:/storage2/fs1/RIS-Services/Active/AlphaFold3,/storage1/fs1/allocation_name/Active:/storage1/fs1/allocation_name/Active /bin/bash

*allocation_name is the name of the user’s Storage allocation. Please DO NOT FORGET to replace the attributes in the above examples with appropriate values carefully.

At the command prompt inside the running job container, the Python virtual environment for the application needs to be activated:
```
. /alphafold3_venv/bin/activate
```
At this point the AlphaFold command is ready to run. If the RIS Service storage space for input data described above is mounted into the container along with a destination for the output data, then the defaults can be used. In this case the only required option is for the output destination:
```
python3 /app/alphafold/run_alphafold.py --output_dir=/path/to/output/dir
```

Batch Job (recommended):

export PATH="/usr/local/cuda/bin/:/opt/conda/bin:/app/alphafold:$PATH" 
export ALPHAFOLD_BASE_DIR=/app/alphafold 
export RIS_SERVICES=/storage2/fs1/RIS-Services/Active/AlphaFold3 
export OUTPUT_DIRECTORY=/storage1/fs1/allocation_name/Active/alphafold3/output 
sbatch job.sh

Job.sh:

#SBATCH --job-name=alphafold_job # Job name
#SBATCH --output=output_%j.log   # Output file
#SBATCH --error=error_%j.log     # Error file
#SBATCH --ntasks=1               # Number of tasks (adjust if needed)
#SBATCH --account=compute2-ris   # Your compute account
#SBATCH --time=01:00:00          # Time limit (hh:mm:ss)
#SBATCH --partition=general-gpu  # Partition/queue to submit to
#SBATCH --gres=gpu:1             # Number of GPUs (adjust if needed)
#SBATCH --mem=4G                 # Memory limit (adjust if needed)
#SBATCH --container-image=ghcr.io#washu-it-ris/alphafold3:2
#SBATCH --container-mounts=/storage2/fs1/RIS-Services/Active/AlphaFold3:/storage2/fs1/RIS-Services/Active/AlphaFold3,/storage1/fs1/allocation_name/Active:/storage1/fs1/allocation_name/Active
. /alphafold3_venv/bin/activate
python3 /app/alphafold/run_alphafold.py --output_dir=/path/to/output/dir