Creativity at Our Fingertips: Real-time, Exhaustive Enumeration of Hyper-scalable Chemical Space with AI

Key Takeaways: 

• We have developed technology that efficiently applies predictive ML models to combinatorial synthesis libraries (CSLs) at a rate of ~500 B compounds per second, per GPU, allowing us to exhaustively evaluate otherwise enumerable chemical space to identify diverse compounds that best fit multi-parameter optimization criteria.
• In practice, this gives chemists the opportunity to flexibly and iteratively explore enormous chemical space while receiving results, in the form of enumerated chemical structures, within seconds of their inquiry, fostering a genuine creative process that leads to higher chemical diversity and developability.
• We use this approach in conjunction with our next-generation extrapolative ML models, which has allowed us to discover high-quality, diverse, structurally-distinct binders for targets of interest; however, this approach can be extended to any advanced predictive model in order to exhaustively evaluate any CSL.