MolEncoder

Molecular AutoEncoder in PyTorch

Install

$ git clone https://github.com/cxhernandez/molencoder.git && cd molencoder
$ python setup.py install

Download Dataset

$ molencoder download --dataset chembl22

Train

$ molencoder train --dataset data/chembl22.h5

Add --cuda flag to enable CUDA. Add --cont to continue training a model from a checkpoint file.

Pre-Trained Model

A pre-trained reference model is available in the ref/ directory. Currently, it performs with ~98% accuracy on the validation set after 100 epochs of training. However, if you succeed at training a better model, feel free to submit a pull request!

TODO

• Implement encoder
• Implement decoder
• Add download command
• Add train command
• Add encode command
• Add decode command
• Add pre-trained model

Shoutouts

• Original paper by Gómez-Bombarelli, et al.
• keras-molecules by Max Hodak
• DeepChem
• PyTorch