We seem to be a little further along and stable than this repo. We are also based on Tensorflow instead of raw cuda which means users can play around with the networks more easily.
Re:Chainer Chemistry: A Library for Deep Learning in Biology...
by brilee on Monday December 18, 2017 @09:12AM
(#15951677)
Excellent. I'd been wanting to implement something like this but these guys beat me to the punch.
Most existing research on NNs in chemistry uses the SMILES string representation and feeds it into a recurrent neural network. The problem with SMILES in my opinion, is that SMILES is not unique, and there are a gazillion ways to represent the same molecule. Thus, the RNN is probably overfitting to the quirks of the SMILES serialization algorithm used by the US Patent Office and/or ChemDraw and/or whatever program was used to generate the library of molecules.
Convolutional graph networks are the obviously correct way to handle molecules.
Re:Chainer Chemistry: A Library for Deep Learning in Biology...
by Pulcinella on Monday December 18, 2017 @09:09AM
(#15951663)
This seems like the development of an artificial chemical intuition[0][1] except since its a computer it can intuit concrete numbers much more comfortably than humans can. I do wonder how its accuracy compares to just running the DFT calculations.
https://deepchem.io/
https://github.com/deepchem/deepchem
http://moleculenet.ai/
We seem to be a little further along and stable than this repo. We are also based on Tensorflow instead of raw cuda which means users can play around with the networks more easily.
Most existing research on NNs in chemistry uses the SMILES string representation and feeds it into a recurrent neural network. The problem with SMILES in my opinion, is that SMILES is not unique, and there are a gazillion ways to represent the same molecule. Thus, the RNN is probably overfitting to the quirks of the SMILES serialization algorithm used by the US Patent Office and/or ChemDraw and/or whatever program was used to generate the library of molecules.
Convolutional graph networks are the obviously correct way to handle molecules.
[0]wavefunction.fieldofscience.com/2016/09/what-is-chemical-intuition.html
[1]https://news.ycombinator.com/item?id=12624036
https://arxiv.org/abs/1610.08935 https://github.com/deepchem/deepchem/blob/master/deepchem/mo...
http://aip.scitation.org/doi/pdf/10.1063/1.4986081
It is a pretty straight-forward application of ML to the space that is starting to have solid results in practice.
- blog: https://preferredresearch.jp/2017/12/18/chainer-chemistry-be...
- github: https://github.com/pfnet-research/chainer-chemistry
- document: http://chainer-chemistry.readthedocs.io/en/latest/index.html