Has anybody installed Alphafold 2 on a cluster for all users? If so, how did you do the install without using Docker? (We don’t use Docker on our cluster due to security issues.) I tried doing a non-Docker install and it’s not working properly.
We are supporting it here at UW-Milwaukee but using Singularity. We followed Open Science Grid’s how-to guide on the installation of Singularity within our HPC environment: Install Singularity - OSG Site Documentation
ComputeCanada had an easy to follow guide on the use of AlphaFold once you have the singularity .sif file created and the appropriate databases downloaded/installed: AlphaFold - CC Doc
Thank you so much for this information!
Yes, I ported their Docker stuff to Singularity. I’ve also modified their run_docker.py script to Singularity, and to integrate with Slurm.
I have a pull request out using my “sing-absl” branch: Add Singularity build recipe and run script. by prehensilecode · Pull Request #182 · deepmind/alphafold · GitHub
My modifications to deal with Slurm are in my “urcf” branch: GitHub - prehensilecode/alphafold at urcf
In both branches, my modifications are limited to the files in a new subdirectory “singularity” of the top level source dir. In the “urcf” branch, there is an example Slurm job script.
Slurm integration includes:
- reading environment variable TMPDIR for temporary folder; if you have set up Slurm to have a job-specific temporary directory TMPDIR, AlphaFold’s output will go there
- reading environment variable SLURM_JOB_GPUS to pass to run_singularity.py