Advancing Machine Learning in Chemistry Through Open and Large-Scale Data Initiatives

Advancing Machine Learning in Chemistry Through Open and Large-Scale Data Initiatives Christoph Steinbeck 15 March 2016: Lee Sedol, a top-ranked Go player, plays against AlphaGo Source: AlphaGo movie The Game of Go The final frontier of human ingenuity • Played since thousands of years • Black and white stones on a 19x19 board • More valid board positions than atoms in the known universe AlphaGo Zero: Mastering the Game of Go without Human Knowledge • DeepMind's AlphaGo Zero: Monte Carlo tree search with a convolutional neural network providing position evaluation and policy guidance. • With only the rules of Go known, AlphaGo Zero improved to superhuman playing strength after a day of training (5 mio games). • Rediscovered 2000 years of Go Theory and expanded it greatly. • Redefined how Go is played today. • It uses just a single machine in the Google Cloud with 4 TPUs Silver, D., Schrittwieser, J., Simonyan, K. et al. Mastering the game of Go without human knowledge. Nature 550, 354–359 (2017). https://doi.org/10.1038/nature24270 Deep neural network can develop superhuman capabilities … when presented with •a well-defined problem and •a lot of training data Big data enables NP cheminformatics Computer-Assisted Structure Elucidation (CASE) CASE applications LUCY (Steinbeck) LSD (Nuzillard) SENECA (Steinbeck) Sherlock (Wenk, Nuzillard, Steinbeck) Part of NMRShiftDB‘s HOSE code table OO(C,C/=&C,=&C/=C,=C) 99.45 OO(C,C/=&C,=&C/=C,=C) 99.56 OO(C,C/=&C,=&C/=C,=C) 99.91 OO(C,C/=&C,=&C/=C,=C) 99.97 OO(C,C/=&C,=&C/=C,=C) 100.47 OO(C,C/=&C,=&C/=C,=C) 101.78 OO(C,C/=&C,=&C/=C,=CO) 101.4 OO(C,C/=&C,=&C/=CC,=CC) 102.3 OO(C,C/=&C,=&C/=CO,=C) 100.7 OO(C,C/=C&,=C&/C,C) 100.7 OO(C,C/=C&,=C&/C,C) 101.2 OO(C,C/=C&,=C&/C,C) 101.3 OO(C,C/=C&,=C&/C,C) 101.6 OO(C,C/=C&,=C&/C,CO) 101.0 O Cl NH 2 *C*CC(*C,*C,=OC/*CX,*&,,CC/&C,,CN,C)C-arom; NMR chemical shift prediction • HOSE codes • Machine Learning • Better and better with more data Hierarchically Ordered Spherical description of Environment (HOSE) Steinbeck, C., & Kuhn, S. (2004). NMRShiftDB … Phytochemistry, 65(19), 2711–2717. https://doi.org/10.1016/j.phytochem.2004.08.027 Big data enables NP cheminformatics Natural Product Likeness (NP-Likeness) • What chemistry can species do or not? Sorokina, M., & Steinbeck, C. (2019). NaPLeS: a natural products likeness scorer—web application and database. Journal of Cheminformatics, 11(1), 55. https://naples.naturalproducts.net Ertl, P., Roggo, S., & Schuffenhauer, A. (2008). J. Chem. Inf. Model, 48(1), 68–74. Big data enables NP cheminformatics 20 most frequent scaffold network scaffolds of COCONUT with their numbers of origin molecules Schaub, J., Zander, J., Zielesny, A., & Steinbeck, C. (2022). Journal of Cheminformatics, 14(1), 79. https://doi.org/10.1186/s13321-022-00656-x Molecular scaffolds and scaffold trees • What chemistry do species or (and not)? Schuffenhauer, A., Ertl, P., Roggo, S., Wetzel, S., Koch, M. A., & Waldmann, H. (2007). J. ChemInform, 38(15), https://doi.org/10.1002/chin.200715213 Big data enables NP cheminformatics RanDepict Brinkhaus, H.O., Rajan, K., Zielesny, A. et al. RanDepict: Random chemical structure depiction generator. J Cheminform 14, 31 (2022). https://doi.org/10.1186/s13321-022-00609-4 Distortion features controlled by fingerprints How to get to big data in NP chemistry? Solution 2: Scientists submit their original data directly to scientific databases and repositories https://nmrxiv.org Caffeine Caffeine - Spectra Chemotion – ELN & Repository 22 Experiment/ Data Collection Experiment Design Data  Processing Analysis Disclosure/ Publication Re-use Acknowledgments Acknowledgments