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.
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.