Category: Data Analysis and Informatics
Early-stage drug discovery requires a constant supply of new molecules, to be fed into High Throughput Screening machines. To increase this supply, virtual molecules can be generated on-demand with neural networks. In this talk, I present a Reinforcement Learning generative model, and a variant using Generative Adversarial Networks. I also present two challenges that both are facing: 1. multitasking between different objectives and 2. generating chemically diverse molecules. Finally, I sketch how these generative models could become a useful proof-of-work for a 'Drugcoin' crypto-currency, in place of the Hashcash proof-of-work of Bitcoin.
Mostapha Benhenda– Researcher, Startcrowd, Châtillon, Ile-de-France, France