From owner-chemistry@ccl.net Sun Feb 26 18:29:00 2017
From: "Thomas Bergwinkl thomas.bergwinkl _ zazuko.com" <owner-chemistry||server.ccl.net>
To: CCL
Subject: CCL: Ligand binding affinity prediction using deep learning
Message-Id: <-52658-170226164051-18847-fFO4PT1QD6KTp5Ly5pybVw||server.ccl.net>
X-Original-From: Thomas Bergwinkl <thomas.bergwinkl!=!zazuko.com>
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=utf-8
Date: Sun, 26 Feb 2017 22:40:45 +0100
MIME-Version: 1.0


Sent to CCL by: Thomas Bergwinkl [thomas.bergwinkl[*]zazuko.com]
Dear all,

I've implemented ligand binding affinity prediction using deep learning
and also wrote a blog post about it. The post doesn't cover only the
technical stuff, but also a little bit the history of the idea:

https://www.bergnet.org/2017/02/ligand-binding-deep-learning/

It uses SMILES token to feed the neural network with the molecule
structures. I'm using alternative representations for data augmentation,
which wasn't used by similar projects before. The training is done with
Keras Gaia, which was developed for this project. It simplifies the
training process for different models and datasets. The SPARQL wrapper
for BindingDB could be of interest for anyone who would like to fetch
BindingDB data using ChEMBL queries. I also implemented a SMILES parser
in JavaScript, because there wasn't any implementation available.

Feedback is welcome!

Regards,
Thomas