In theory, there are more potential stable molecules than there are atoms in the universe. So how do we find new molecules for the drugs, materials, additives (etc) of the future? High-throughput virtual screening now allows us to scan millions of molecules, but this is still many orders of magnitude too small. The use of deep learning can both accelerate our search by speeding up (for example) property calculations, and by moving us towards an intelligent means of searching this vast space. I will introduce various types of neural network, and show how you can use them to speed up search problems by coupling them with a state of the art technique called Bayesian Optimization.
YOU MAY ALSO LIKE:
Using Deep Learning to Design the Molecules of Tomorrow
Edward Pyzer-Knapp obtained his PhD from the University of Cambridge using simulation-based techniques to help guide experimental materials design. He then moved to Harvard where he was in charge of the day-to-day running of the Harvard Clean Energy project, which was a collaboration with IBM which combined massive distributed computing, quantum-mechanical simulations, and machine-learning to accelerate discovery of the next generation of organic photovoltaic materials. Now, Edward works for IBM Research UK, leading the large-scale machine learning research effort with a specific interest in bringing cutting edge techniques from machine learning and data science to the physical and life sciences.