Automating Chemical Synthesis using AI and Automated Systems

P. Madrid, J.P. Malerich, M. Latendresse, M. Krummenacker, D. Stout, J-P. Lim, Vi-Anh Vu, J. Szeto, K. Rucker, J. White, N. Collins
SRI,
United States

Keywords: AI, chemical synthesis

Summary:

Artificial intelligence can greatly increase the speed and accuracy of molecular design in drug discovery. However, computationally designed molecules still need to be synthesizable and synthesized on a scale suitable for testing—both of which require significant human intervention. SRI has created the SynFini suite of tools to automate the process from route design through the multi-step synthesis of small molecule drug-like compounds. One part of SynFini is a synthetic planning tool, SynRoute, that combines rapid search methodologies for large reaction databases with computer-generated reactions based on machine learning classifiers. Classifiers for common reaction transformations have been built using both Random Forrest and Multilayer Perceptron methods to achieve on average about 91% accuracy. Novel reaction or reactions with missing data from literature references can be rapidly tested and optimized using an automated reaction printing system, SynJet. This robotic system uses thermal inkjet printing technology to rapidly dispense reagents, then process individual reaction vials at prescribed times and temperatures based on statistical design of experiment methodologies. The completed reactions are then processed by analytical instrumentation to produce reliable data on the feasibility and optimal conditions for chemical reactions. Finally, multi-step synthetic routes for target compounds or intermediates are performed on an automated continuous flow multistep chemical synthesizer called AutoSyn that makes milligram to gram-scale amounts of virtually any drug-like small molecule in a matter of hours. Characterization and monitoring of reaction output are measured continuously by on-line LC-MS and in-line 1H-NMR. The automated captures of high-quality synthesis data is recorded and will be used to continuously improve future synthetic route planning and synthetic process optimization. Moreover, our approach enables digital synthesis protocols that ensure reproducibility and transferability of synthesis procedures. Examples of the application of AI synthetic planning combined with automated synthesis will be presented to demonstrate the utility of applying AI and advanced analytics to the process of chemical synthesis.