Deep convolutional neural networks (neural nets with a constrained architecture that leverages the spatial and temporal structure of the domain they model) achieve the best predictive performance in areas such as speech and image recognition. Such neural networks autonomously discover and hierarchically compose simple local features into complex models.
Biochemical interactions, being similarly local, are amenable to automatic discovery and modeling by similarly constrained machine-learning architectures. Abe Heifets offers an overview of AtomNet, a structure-based deep convolutional neural network designed to predict the bioactivity of small molecules for drug discovery applications. Abe discusses training AtomNet on millions of training examples derived from ChEMBL and the PDB and visualizes the automatically derived convolutional filters, demonstrating that the system is discovering chemically sensible interactions. Abe concludes by showing how autonomously discovered filters can outperform previous docking approaches and achieve an AUC greater than 0.9 on 57.8% of the targets in the DUDE benchmark and that AtomNet’s application of local convolutional filters to structural target information successfully predicts new active molecules for targets with no previously known modulators.
Abraham Heifets is the cofounder and CEO of Atomwise, which uses machine learning to help discover new medicines. Previously, Abe researched high-performance data processing at IBM’s T.J. Watson Research Center and helped develop the strategy and control AI system of the world-champion robotic soccer team at Cornell University. He created SCRIPDB, one of the largest public databases of patented chemical structures at the time, and LigAlign, a protein analysis tool used by researchers in 70 countries. He is an author on 19 papers, patents, and patent applications and was named Time magazine’s person of the year in 2006. Abe was a Massey fellow at the University of Toronto and a fellow of the Ontario Brain Institute. His doctoral work applied machine learning and classical artificial intelligence techniques to organic synthesis planning, a long-standing challenge in chemistry.
©2017, O'Reilly Media, Inc. • (800) 889-8969 or (707) 827-7019 • Monday-Friday 7:30am-5pm PT • All trademarks and registered trademarks appearing on oreilly.com are the property of their respective owners. • firstname.lastname@example.org