AI is a viable alternative to high throughput screening: a 318-target study
The Atomwise AIMS Program, Izhar Wallach, Denzil Bernard, Kong T. Nguyen, Gregory Ho, Adrian R. Morrison, Adrian Stecuła, Andreana M. Rosnik, Ann O’Sullivan, Aram Davtyan, Ben Samudio, Bill Thomas, Brad Worley, Brittany Butler, Christian Laggner, Desiree A. Thayer, Ehsan Moharreri, Greg Friedland, Ha H. Truong, Henry van den Bedem, Ho Leung Ng, Kate A. Stafford, Krishna K. Sarangapani, Kyle E. Giesler, Lien Ngo, Michael M. Mysinger, Mostafa Ahmed, Nicholas J. Anthis, Niel M. Henriksen, Paweł Gniewek, S.R. Eckert, Saulo de Oliveira, Shabbir Suterwala, Srimukh Veccham Krishna PrasadPrasad, Stefani Shek, Stephanie Contreras, Stephanie R. Hare, Teresa A. Palazzo, Terrence E. O’Brien, Tessa Van Grack, Tiffany R. Williams, Ting‐Rong Chern, Victor Kenyon, Andreia H. Lee, Andrew B. Cann, Bastiaan Bergman, Brandon Anderson, Bryan D. Cox, Jeffrey M. Warrington, Jon M. Sorenson, Joshua M. Goldenberg, Matthew A. Young, Nicholas DeHaan, Ryan P. Pemberton, Stefan Schroedl, Tigran M. Abramyan, T. Raghavendra Gupta, Venkatesh Mysore, Adam Presser, Adolfo A. Ferrando, Adriano D. Andricopulo, Agnidipta Ghosh, Aicha Gharbi Ayachi, Aisha Mushtaq, Ala M. Shaqra, Alan Kie Leong Toh, Alan V. Smrcka, Alberto Ciccia, Aldo Sena de Oliveira, Aleksandr Sverzhinsky, Alessandra Mara de Sousa, Alexander I. Agoulnik, Alexander Kushnir, Alexander N. Freiberg, Alexander V. Statsyuk, Alexandre R. Gingras, Alexei Degterev, Alexey Tomilov, Alice Vrielink, Alisa A. Garaeva, Amanda Bryant-Friedrich, Amedeo Caflisch, Amit K. Patel, Amith Vikram Rangarajan, An Matheeussen, Andrea Battistoni, Andrea Caporali, Andrea Chini, Andrea Ilari, Andrea Mattevi, Andrea Foote, Andrea Trabocchi, Andreas Stahl, Andrew B. Herr, Andrew D. Berti, Andrew Freywald, Andrew G. Reidenbach, Andrew Lam, Andrew Cuddihy, Andrew D. White
Abstract
High throughput screening (HTS) is routinely used to identify bioactive small molecules. This requires physical compounds, which limits coverage of accessible chemical space. Computational approaches combined with vast on-demand chemical libraries can access far greater chemical space, provided that the predictive accuracy is sufficient to identify useful molecules. Through the largest and most diverse virtual HTS campaign reported to date, comprising 318 individual projects, we demonstrate that our AtomNet® convolutional neural network successfully finds novel hits across every major therapeutic area and protein class. We address historical limitations of computational screening by demonstrating success for target proteins without known binders, high-quality X-ray crystal structures, or manual cherry-picking of compounds. We show that the molecules selected by the AtomNet® model are novel drug-like scaffolds rather than minor modifications to known bioactive compounds. Our empirical results suggest that computational methods can substantially replace HTS as the first step of small-molecule drug discovery.