"Machine learning in chemistry and beyond" (ChE-605) seminar by Prof. Klavs F. Jensen: "Accelerating Chemical Discovery and Development with Machine Learning, Robotics, and Automation"

Machine learning (ML) tools are becoming increasingly effective at generating new candidate molecules, predicting their properties, proposing reaction pathways through computer-aided synthesis planning (CASP), and analyzing analytical data. Automation and robotic technologies have also become easier to use and more affordable to implement, enabling automated chemical synthesis and characterization with little or no human intervention once the system is set up. Case studies demonstrate how automated synthesis systems integrated with ML algorithms create autonomous chemical discovery platforms capable of operating across diverse chemical spaces with minimal manual effort, improving the traditional design-make-test-analyze (DMTA) workflow. The synthesis of new organic dye molecules exemplifies that property–focused discovery platforms can suggest and synthesize molecules to expand training datasets for ML generative and property-prediction models, helping to map the chemical space and ultimately identify top-performing molecules. A second example illustrates how the automated platform can be easily adapted to discover histone deacetylase inhibitors by modifying the underlying ML models and using Bayesian optimization to balance experimental costs with the number of molecules screened during sequential rounds of virtual, coarse, and refined experimentation. A final example, identifying electrochemical oxidation transformations and their optimal reaction conditions, highlights how Large Language Models (LLMs) further facilitate the integration of AI tools into automated chemical experimentation. Challenges, opportunities, and the role of the human operator are discussed in all the case studies.