QBMG: Quasi-Biogenic Molecule Generator with Deep Recurrent Network

Introduction

This is a PyTorch implementation of the paper: QBMG: quasi-biogenic molecule generator with deep recurrent neural network The user can (i) obtain a certain number of unbiased biogenic-like molecules and (ii) obtain a certain number of focused chemotype derivatives with transfer learning.We thank the previous work by the Olivecrona team.The code in this repository is inspired on REINVENT.

sample.py: script that is used to generate a certain number of molecules (both unbiased or certain chemotype biogenic-like molecules) with trained model.

Example:sampling 1000 molecules from unbiased model and focused chemotype model respectively

   %run sample.py ./data/biogenic.ckpt 1000

   %run sample.py ./data/coumarin.ckpt 1000

transfer_learning.py: script that is used to train focused chemotype biogenic moleucles and obtain focused chemotype derivatives. The user can provide own focused chemotype molecules to generate new derivatives.

Example:training a focused chemotype library

   %run transfer_learning.py ./data/coumarin.smi

This package requires:

Name		Name	Last commit message	Last commit date
Latest commit History 20 Commits
data		data
image		image
.gitignore		.gitignore
LICENSE		LICENSE
README.md		README.md
Voc		Voc
data_struct.py		data_struct.py
model.py		model.py
sample.py		sample.py
train.py		train.py
transfer_learning.py		transfer_learning.py