Running AlphaFold on the Della cluster at Princeton University

This repo contains a SLURM job submission script to run AlphaFold on the Della cluster at Princeton University. This script loads the AlphaFold module and then calls the run_alphafold.sh wrapper, which in turn executes the run_alphafold.py script. These scripts have been kindly written by Matthew Cahn (MOL) and can be found at /tigress/MOLBIO/local/src/alphafold-<version>/run_alphafold.sh. He also has SLURM script examples at example-alphafold-<version>-norelax.cmd

• AlphaFold version: 2.3.1
• Database date: 2022-12-20

Check that you have access to the Della computing cluster and the Della cryo-EM compute nodes

Open a Terminal window on your computer (Mac), wsl window (Windows Subsystem for Linux), or internet browser and go to http://mydella.princeton.edu

• For Terminal or wsl, try to log onto Della: ssh <netid>@della.princeton.edu. If there are no errors, you have access to Della
• For MyDella, click on the tabs at the top for 'Clusters' and then 'Della Cluster Shell Access'. If there are no errors, you have access to Della After connecting to Della, check your group association: groups <netid>. If you see 'cryoem', you likely have access to the compute nodes.

If you don't have access to the Della computing cluster or the Della cryo-EM compute nodes

Email Research Computing: [email protected]

• Send them your name, netid, sponsor/PI
• Ask for Della access if you don't already have it, and provide your primary group affiliation ('molbio' for MOL/Ploss, 'genomics' for LSI/Adamson)
• If you are in Ploss lab, also ask to be added to the 'ploss' group
• Let them know that you will be using AlphaFold, and ask them to add you to the 'cryoem' group and for access to QOS della-cryoem

Dear Research Computing,

I am a <student/postdoc/staff> (<netid>, sonpsor: Alexander Ploss). I will be working with AlphaFold on the Della cluster, so could you please create an account for me on Della, with the primary group 'molbio', additional groups 'ploss' and 'cryoem', and permit me access to the QOS della-cryoem?

Thank you,
<name>

Log onto the Della cluster and create a project directory

1. Connect to the Princeton VPN using GlobalProtect (follow 'Installing GlobalProtect Software': https://workcontinuity.princeton.edu/remoteaccess) even if you are on campus
2. Log onto Della (ssh <netid>@della.princeton.edu for Terminal (Mac) or wsl (Windows Subsystem for Linux) or using http://mydella.princeton.edu
3. Navigate to your personal directory on the scratch volume, typically cd /scratch/gpfs/<netid>
4. Create a new project directory with mkdir <project-directory>; cd <project-directory>
5. Clone this github repo into the directory with git clone https://github.com/09aaronl/alphafold_princeton-della/ .
6. List the contents of your project directory with ls. You should see 6 files:

alphafold.slurm  EXAMPLE_protein1.fa  EXAMPLE_proteins12.fa  EXAMPLE_proteins.txt  LICENSE  README.md

Set up your proteins to fold

6. RECOMMENDED: As a test run, you can use the EXAMPLE files provided. Enter the command cp EXAMPLE_proteins.txt proteins.txt and then SKIP steps 7 and 8 until the test run completes successfully.

7. Using a text editor like nano or vim, create a text file named proteins.txt, containing one file name per line, a comma, and the mode to use. Names can be individual proteins ('monomer' mode) or multiple proteins to co-fold together ('multimer' mode). For example:

protein1.fa,monomer
protein2.fa,monomer
proteins12.fa,multimer
...
proteinN.fa,monomer

8. For each fold, create a FASTA file matching the file name in proteins.txt. EMBOSS has a nice tools for converting text to FASTA format: https://www.ebi.ac.uk/jdispatcher/sfc/emboss_seqret

• A monomer (protein1.fa) should look like:

>protein1
MPRTEINSEQWITHTESTWRAPPINGAFTEREIGHTYCHARACTERSPRTEINSEQWITH
TESTWRAPPINGAFTEREIGHTYCHARACTERSPRTEINSEQWITHTESTWRAPPINGAF
TEREIGHTYCHARACTERS

• For multimers (proteins12.fa), add all FASTA headers and sequences in the same file:

>protein1
MPRTEINSEQWITHTESTWRAPPINGAFTEREIGHTYCHARACTERSPRTEINSEQWITH
TESTWRAPPINGAFTEREIGHTYCHARACTERSPRTEINSEQWITHTESTWRAPPINGAF
TEREIGHTYCHARACTERS
>protein2
MSECNDPRTEINSEQFRCFLDINGSECNDPRTEINSEQFRCFLDINGSECNDPRTEINSE
QFRCFLDINGSECNDPRTEINSEQFRCFLDING

Edit the SLURM job submission script

9. Edit the file alphafold.slurm with a text editor:

• Change the line #SBATCH --mail-user=<netid>@princeton.edu to include your netid
• Change the line #SBATCH --time=06:00:00 to the maximum expected runtime, typically 1 h per 250 residues
  • This does not need to be edited for the test run with the EXAMPLE files
• Change the line #SBATCH --array=1-2 to specify which lines to fold from the proteins.txt file, 1-indexed, inclusive
  • This does not need to be edited for the test run with the EXAMPLE files

10. Submit the job with sbatch alphafold.slurm and you can monitor the job status periodically with squeue --me
11. When the job ends due to failure or success, you will receive an email notification, and squeue --me should return nothing
12. List the contents of your project directory with ls. You should see a new folder for each fold, containing PDB files with your predicted structures!