Structure-based virtual screening of ultra-large chemical spaces: Advances and pitfalls

On-demand chemical spaces consist of molecules that are, a priori, readily synthesizable from sets of commercial building blocks through robust organic reactions. As these spaces expand-now reaching the scale of several trillions of compounds-computational chemists are compelled to develop innovative algorithms for efficient enumeration, storage, and virtual screening, particularly when three-dimensional constraints of target proteins are involved. This review examines the primary approaches to structure-based ultra-large virtual screening, highlighting the significant advantages of screening at such a scale while addressing the remaining practical and theoretical hurdles. Current prospective applications, often relying on brute-force docking, typically report improved hit rates and more potent primary hits; however, they must contend with the exponential growth of available chemical space. To address this, recent developments have integrated active learning, probabilistic sampling, and synthon-guided methods to accelerate docking and prioritize the most promising compounds. Finally, we provide a perspective on the transformative impact of ultra-large chemical spaces on early hit identification and the overall organization of early drug discovery.