PL_Binding
Task: For each protein molecule, predict the top 10 ligands with the highest binding affinity.
The model training is framed as a binary classification problem, according to whether a protein-ligand input pair binds or not.
Raw data are coordinate features and atomic type. The pre-processing converts these to a 3D voxel for input into a 3D CNN.
Execute the jupyter notebook in order, for training and evaluation.
Files
- dataset.py: preprocessing, dataloading, transformations
- train.ipynb: notebook to train CNN model
- predict.ipynb: notebook to run evaluation on the test data. requires a trained model.
- model (folder): pre-trained models can be found here
- data (folder): some toy data
Dependencies
- pytorch 0.4.0 +
Authors
- Muhammad Huzaifah
- Luis Vasquez
This project was initially written for CS5242 Neural Networks and Deep Learning project offered by the National University of Singapore. Original data was prepared by the course administrators.