Deep Recurrent Networks for Molecular Drug Design
Authors
Abstract
In drug discovery, deep learning algorithms have emerged to be an effectivemethod to generate novel chemical structures. They can speed up this
process and decrease expenditure. We propose a computational model
for molecular de novo drug design that is able to produce new drug compounds.
This computational model based on the recurrent neural network
(RNN) can learn the syntax of molecular representation in terms of Simplified
Molecular Input Line Entry Specification (SMILES) strings. The
model and its generated SMILES are evaluated using MolVS tool syntactically
and biochemically. We analyze the best recurrent network and
the parameters. The network that reaches the best result, 98% of valid
SMILES, was an RNN containing long short term memory(LSTM) cells.