NeuralPlexer workflow

Predicting the interaction between ligands and proteins with NeuralPlexer, a deeplearning tool by Qiao et al. 2024.

Installation

The following commands install dependencies of the workflow in the current directory within the .pixi folder. After installation, you cannot move the folder without re-installling all the dependencies.

curl -fsSL https://pixi.sh/install.sh | bash
# ... cd <this repo>
pixi install

Usage

Each job has a json descriptor in the data/ folder:

[
  {
    "name": "2024-05-13_20:16",
    "parameters": {
      "sampler":"langevin_simulated_annealing",
      "n-samples": 10,
      "chunk-size": 1,
      "num-steps": 40
    },
    "ligands" : [
       {
        "molecule": {
          "sdf": "data/lig_ref.sdf",
          "count": 2
        }
      }
    ],
    "sequences": [
      {
        "proteinChain": {
          "sequence": "FGGGFGGGGGSGSGSGG",
          "count": 2
        }
      },
      {
        "proteinChain": {
          "sequence": "FGGSGSGSGG",
          "count": 1
        }
      }
    ]
  }
]

The ligands are included like this.

data
├── lig_ref.sdf
└── test.json

This will make a folder in results/data/<jobname> with each job in it. Right now, the token limit seems to be 600 amino acid residues. At this point, the graphical card runs out of memory.

• pixi run make runs the full workflow. You can supply arguments to snakemake as needed, such as --cores 10, if your process needs 10 cores.

• pixi run test runs a dry run (-np) of the workflow.

• pixi run update_dag updates the directed acyclic graph of the snakemake workflow and moves it to the resources folder.

• pixi run help show the usage of the Snakefile

Usage in a HPC context is done with the

pixi run slurm

command, this will launch the hardware intensive snakemake tasks as SLURM jobs.

About

When pixi run make is executed, the snakemake pipeline sets up neuralplexer and predicts each listed complex in data. Additionally, summary statistics are calculated and plotted as shown below. These summary statistics end up in results/analysis. Statistics and analysis include:

• ramachandran plot.

• residue wise plddt.

• sasa surface Accessibility.

• Number of steric clashes.

• Identification of interacting residues using plip.