run_masif

• Everything should be setup to run on alpine.

  • Masif neosurf container link

• You'll need to copy the run_masif_batch.sh script from my projects directory:

  cp /projects/liwh2139/masif/run_masif_batch.sh </path/where/you/want/it/>

• What the submission script contains:

  Slurm directives:

    #SBATCH --account=ucb-general
    #SBATCH --job-name=masif-preprocess
    #SBATCH --partition=atesting        
    #SBATCH --nodes=1
    #SBATCH --time=01:00:00		    
    #SBATCH --output=/scratch/alpine/%u/masif_runs/%x_%j.out
    #SBATCH --error=/scratch/alpine/%u/masif_runs/%x_%j.err
  

  • You'll want to change the partition to amilan and the time will depend on how many structures you'll process (it takes ~2-3 minutes). It is currently set to run on the atesting partition for 1 hr.
    • If working with really large datasets, you'll want to add #SBATCH --qos long for jobs that will take longer than 24hrs

  Modules:

    # Load required modules
    module purge
    module load singularity
  

• Clears out anything that may have previously been loaded and loads singularity to run the .sif container

  Conda:

   # source your conda env
   # conda activate env_name
  

• The script requires python, so any environment you have that has some version of python 3 will work. Just uncomment the conda activte line and list your environment name in place of env_name.

  Directory organization:

  # Set directories
  if [ -z "$1" ]; then
    echo "[ERROR] You must provide the PDB directory as the first argument."
    echo "Usage: sbatch run_masif_batch.slurm <pdb_dir> [base_dir]"
    exit 1
  fi
  
  PDB_DIR=$(realpath "$1")
  SCRATCH_DIR="/scratch/alpine/$USER"
  BASE_DIR=${2:-"$SCRATCH_DIR/masif_runs"}
  OUT_DIR="$BASE_DIR/masif_output"
  LOG_DIR="$BASE_DIR/logs"
  SIF_PATH="/projects/liwh2139/masif/masif-neosurf.sif"
  
  # Make sure output folders exist
  mkdir -p "$OUT_DIR" "$LOG_DIR"
  
  # Echo paths for confirmation
  echo "PDB directory:     $PDB_DIR"
  echo "Output directory: $OUT_DIR"
  echo "Log directory:    $LOG_DIR"
  echo "SIF path:         $SIF_PATH"
  

  • The only mandatory thing you have to declare is the path to the directory where you have the structures/pdbs. When submitting to slurm you run

    sbatch run_masif_batch.sh <path/to/structs/directory/

  • You could also declare the output and log dirs but the default is set to write to scratch. It will create a parent directory called masif_runs and within that directory you'll find all the files.

  Running the python wrapper:

  # Run the wrapper Python script
  python3 /projects/liwh2139/masif/run_masif_sif_v2.py \
    --pdb_dir "$PDB_DIR" \
    --output_dir "$OUT_DIR" \
    --log_dir "$LOG_DIR" \
    --sif "$SIF_PATH"
  

  • The script will take the pdb directory and loop through and run all the structures through the precompute stages of the masif neosurf protocol.
    • masif only accepts files with 4-digit ids/pdb like codes. The script will simlink the input pdbs and mask the original names with 4-digit numerical codes. It is set up to run on chain A of the protein.

• Other info for pulling npy

  • Currently set to fill missing values/NaNs with 0s.

  • extract_descriptors_to_csv.py can take in a supplementary metadata file that should be of the format:

    • structure,sequence,ligand_name,ligand_family

  • The structure column must match the original structure names in pdb_mapping.txt.

  • Additional columns (e.g., sequence, ligand_name, ligand_family) are optional and customizable.

  • Metadata will be merged into the final descriptors CSV to enable downstream visualization (e.g., coloring t-SNE plots by ligand family).

  • The t-sne script will take in that csv and output a png and an additional csv with point coordinates.