Predicting multiple protein structures using Alphafold on Notos

Setup

1. Copy the alphafold folder from /mnt/beegfs/alphafold/alphafold to your home directory
   • When you log in to notos you are automatically in your home directory
   • Run the command below to copy the folder into your home directory
   • cp -r /mnt/beegfs/alphafold/alphafold/ .
   • This folder has all the required updates to run alphafold successfully
2. Locate the folder named docker by running the command cd alphafold/docker
3. Open run_docker.py by running the command vi run_docker.py
   • Press i to insert and make edits to the file
   • Change the destination folder output_dir on line 49 to a folder you wish the results are saved into
   • Make sure you have sufficient read and write permissions for the output_dir folder

Running multiple files on Alphafold in series on a single GPU

1. Create a folder with all the sequences you want to run on a single GPU
2. Check the status of the GPU’s by running the command nvidia-smi
3. cd into your home directory using the command cd
4. Locate and enter the folder GPU_scripts by running the command cd alphafold/GPU_scripts
5. List contents in the folder by using the command ls. The folders are named GPU_[GPU_number].sh. Choose the GPU number which isn’t being used (Following step 2)
6. Start a named screen session, by typing the following in your console screen -S session_name

7. Run the following command to run the sequences on a single GPU

   bash GPU_[GPU_number].sh path_to_the_folder_with_sequences

   For example, if you want to run the sequences on GPU number 6: bash GPU_6.sh ~\sequences\test_folder\ where ~\sequences\test_folder\ is the folder with sequences.

   • These scripts run alphafold using full databases. For optimized performance use the preset reduced_dbs. Open the GPU_[number] script and add the preset --db_preset=reduced_dbs to line 9
   • Other presets and options can be found here
8. Detach from the linux screen session, by press Ctrl+a d
   • To get the list of current running sessions, type the following screen -ls
   • To reattach to a linux screen, type the following screen -r session_name

Running multiple files on Alphafold in series on a multiple GPUs

1. Create seperate folders for each set of sequences you want to run on the GPUs
2. Repeat steps 2 to 8 for each of the folders on seperate GPUs