Are you an EPFL student looking for a semester project?
Work with us on data science and visualisation projects, and deploy your project as an app on top of Graph Search.
This lecture covers the use of generative models and property predictions in accelerating organic synthesis with chemical language models. It explains the methodology of representing molecules as text using SMILES, a chemical notation system. The lecture also delves into the basics of SMILES, semantic constraints, and the future of molecular string representations. Furthermore, it discusses the design, test, and synthesis planning of molecules, as well as the application of sequence-to-sequence models and the Transformer architecture in chemical language processing. The Molecular Transformer model for uncertainty-calibrated chemical reaction prediction is presented, highlighting its accuracy on unseen reactions and its superiority over rule-based approaches. Additionally, the lecture explores the separation vs. mixed setting in chemical reactions, human prediction benchmarks, and the significance of stereochemistry and experimental validation in molecular transformations.