Classification of Drug Like molecule using Neural Networks.
more about DrugAI.. http://gananath.github.io/drugai.html
-
Python 2.7
-
Keras(Theano/Tensorflow)
-
Pandas
-
Scikit-Learn
Generator script for creating drug like molecule using LSTM model. Read more from here http://gananath.github.io/drugai-gen.html
This is my own experiments with Generative Adverserial Network (GAN) for drug like molecule generation. Teaching GAN in discrete dataset is hard and also I learned to code GAN from internet so would not gurantee any acurracy of the results or the code. Read more from here http://gananath.github.io/drugai-gan.html
A Wasserstein GAN model with CNN; this model currently trains the fastest and probably gives the best result.
# Samples Generated
['CC1=C(C(C(=O)O)(=CC=N2[S]CCCCCC(C(Cl)C1C4)[+])C2=C4=O|||||||||||||||||||||||||||' 'CC1=C(C(C(=O)OO(=CC=N2[N]CCCCCC(C(Cl)C1C3)[+])C2=CC=O)||||||||||||||||||||||||||' 'CC1=C(C(C(=O)O)(=CC=N2[N]CCC=CC(C(CO)C1C3)[+])C2=CC=O)||||||||||||||||||||||||||']
- Recently I came across GAN's which uses condition like cGAN,acGAN etc. which uses a noise+class as input whereas I am using only class.
Because I seen a increase in interest for DrugAI-Gen.py; for programmers I have added another dataset sms.tsv. It contains SMS spams. Try to use it for generating Spam's and Ham's.
@misc{gananath2016,
author = {Gananath, R.},
title = {DrugAI},
year = {2016},
publisher = {GitHub},
journal = {GitHub repository},
howpublished = {\url{https://github.com/Gananath/DrugAI}}
}