LigandNet, a tool which combines different machine learning models into one platform for the prediction of the state of the ligands either actives or inactives for a particular proteins.

Create a conda environment using environment.yml. Run the following

conda env create -f environment.yml

Use ligandnet.py to run predictions. To see the available options, run python ligandnet.py --help which shows the following:

usage: ligandnet.py [-h] [--sdf SDF] [--smiles SMILES]
                    [--confidence CONFIDENCE]

Ligand activity prediction using LigandNet

optional arguments:
  -h, --help            show this help message and exit
  --sdf SDF             SDF file location
  --smiles SMILES       SMILES
  --confidence CONFIDENCE
                        Minimum confidence to consider for prediction. Default
                        is 0.5

For example, python ligandnet.py --smiles CCCC will run all the LigandNet models on the compound CCCC. For an sdf file as input, run python ligandnet.py --sdf samples/AAAAML.xaa.sdf. The parameter confidence is the minimum probability for which a model will consider a ligand as an active.

To get the decoys used for training the LigandNet models, run

1. bash get_decoys.sh
2. tar xvf decoys.tar.gz

An web interface for ligand activity prediction using the LigandNet models is available at LigandNet