Base2Backbone

Setup

1. Clone the repository:

git clone https://github.com/Eclipter/base2backbone.git
cd base2backbone

1. Install Conda. See Conda Installation
2. Download PyNAMod:

git clone https://github.com/intbio/PyNAMod.git pynamod
cd pynamod

1. Create a new environment with PyNAMod dependencies:

/opt/miniconda/bin/conda env create \
  --name base2backbone \
  --file pynamod/environment.yml

1. Install the project dependencies:

/opt/miniconda/bin/conda env update \
  --name base2backbone \
  --file environment.yml
/opt/miniconda/bin/conda run \
  -n base2backbone \
  pip install \
    -e ./pynamod \
    -e .[test] \
    --no-deps

Training

1. Make sure to set up the environment. See Setup
2. Train and test the model:

python scripts/train.py

1. Analyze the results if necessary

python scripts/analyze.py

1. Export the best model to ONNX:

python scripts/export.py --run-id torsions/1/baseline

Usage

1. Make sure to set up the environment. See Setup
2. Predict the backbone (alternatives are listed below):

• CLI (for a single structure):

  base2backbone \
    --input input.pdb \
    --output output.pdb

  Input format may be PDB or mmCIF. Output may be PDB or mmCIF independently. Both are inferred from the filename extension.

  By default 5'-terminal phosphate atoms (P, OP1, OP2) are not predicted. Pass --generate-5-prime-phosphate to include them.

• CLI (for a trajectory):

  • Single multi-model file output:

    base2backbone \
      --input topology.pdb \
      --trajectory traj.xtc \
      --output output_traj.pdb

    Input format may be PDB or mmCIF. Output may be PDB or mmCIF independently. Both are inferred from the filename extension.

  • Example (one PDB per frame):

    base2backbone \
      --input topology.pdb \
      --trajectory traj.xtc \
      --output-dir frames \
      --output-format .cif

    Input format may be PDB or mmCIF. Output may be PDB or mmCIF independently. Output format is set by the --output-format argument.

• Python API: The library always returns an MDAnalysis.Universe.

  import MDAnalysis as mda
  from base2backbone import predict_backbone, predict_backbone_trajectory
  
  single = predict_backbone('input.cif')
  traj_in = mda.Universe('topology.pdb', 'traj.xtc')
  traj_out = predict_backbone_trajectory(traj_in)