Abstract:

The discovery of functional molecules plays a fundamental role in advancing chemical science and engineering, yet it remains a costly and time-intensive process. Recent advances in computational methods, particularly in generative artificial intelligence, have introduced a new approach—generative molecular design—which holds the promise of efficiently identifying molecules with desired properties. However, despite significant progress, their practical impact in real-world applications has been limited. In this talk, I will present our efforts to address critical bottlenecks in generative molecular design, namely synthetic accessibility and sample efficiency. I will present the development of benchmarks that capture real-world complexity and the development of chemistry-tailored solutions to enhance the practicality of generative algorithms. Taken together, these advances aim to close the gap between computational innovation and practical feasibility, paving the way for the accelerated, AI-driven discovery of novel functional molecules.