Lecture
This lecture discusses the application of generative models in molecular design, focusing on their potential to discover novel molecules that can address significant challenges in chemistry. The instructor begins by highlighting the vast unexplored chemical space, where the number of theoretically feasible molecules far exceeds those that have been synthesized. The lecture covers two main approaches: general molecular generation, which broadly searches for property optima, and candidate-based molecular generation, which refines existing candidates. The instructor introduces the concept of iterative graph-to-graph translation as a method to optimize molecular properties, particularly aqueous solubility. The discussion extends to integrating generative models into autonomous labs, emphasizing the importance of synthesizing feasible molecules. The lecture concludes with insights into the challenges faced in molecular discovery, including the reliance on surrogate models and the need for accurate predictions. Overall, the lecture provides a comprehensive overview of how generative models can revolutionize molecular design and discovery processes.