Review Manuscript: RL-GANs in Drug Discovery: ECAAE & ATNCPosted on 3/17/2019
Generative Adversarial Networks (GANs), the leading type of neural networks are drastically reducing cost and time consumed in de novo drug discovery. Novel molecular structures toke months of work in the conventional pipeline just a few years ago. Now, GANs can do the same work in hours. GANs operate by generating new drugs to target disease, instead of relying on screening from limited drug databases as previous classifiers have attempted. Two GANs have been published as of October 2018: the Entangled Conditional Adversarial Auto-encoder (ECAAE)1 and the Adversarial Threshold Neural Computer (ATNC)2. Both have shown some of the first generated molecule in vitro validation ever from neural networks. Validation for both has shown good IC50, selectivity, and inhibition potency for their respective hit kinase inhibitors. The targeted kinases from the successful inhibitor validation tests were JK3, SGK1, and Aurora A/B. These kinases show hyper phosphorylation in cancer of at least 8 organs, Parkinson’s disease, ALS, rheumatoid arthritis, psoriasis, and vitiligo.
1. Daniil Polykovskiy, Alexander Zhebrak, Dmitry Vetrov, Yan Ivanenkov, Vladimir Aladinskiy, Polina Mamoshina, Marine Bozdaganyan, Alexander Aliper, Alex Zhavoronkov, and Artur Kadurin. Entangled Conditional Adversarial Autoencoder for de Novo Drug Discovery. Molecular Pharmaceutics 2018 15 (10), 4398-4405
2. Evgeny Putin, Arip Asadulaev, Quentin Vanhaelen, Yan Ivanenkov, Anastasia V. Aladinskaya, Alex Aliper, and Alex Zhavoronkov. Adversarial Threshold Neural Computer for Molecular de Novo Design. Molecular Pharmaceutics 2018 15 (10), 4386-4397