The scale of screening within ultra-large chemical spaces plays a pivotal role in contemporary drug discovery, particularly during the initial stages of hit identification. Construction of such chemical spaces to allow their efficient exploration for compound discovery is a critical challenge. In this research, we have generated a novel combinational chemical library, through integration of a curated reaction set with over 1.8 million available building blocks, resulting in a chemical space with more than one trillion molecules theoretically.To characterize the structural diversity, novelty, and physicochemical properties of the compounds in this trillion-scale library, a randomly sampled subset as a surrogate for exploration, rather than an exhaustive search algorithm, was employed. Our results demonstrate that, at both the fragment-level and molecule-level chemical spaces, the constructed library encompasses broad physicochemical diversity and rich scaffold novelty, overlapping with but also extending beyond natural product and FDA-approved chemical spaces. Scaffold retrieval analyses indicated near-ideal structural diversity at scale. Following virtual screening, the hit compounds need to be chemically synthesized, which is often resource demanding. Leveraging the unique characteristics of the combinatorial chemical space and employing a Quadratic Unconstrained Binary Optimization (QUBO) model, we have developed a strategy to maximize the utilization of building blocks for chemical synthesis to generate a larger number of molecules with desired properties (drug-likeness, natural product-likeness, and structure similarity). Together, our work has established a theoretically trillion-scale combinatorial chemical library which can facilitates efficient virtual screening and hit identification and further developed a novel method for resource-optimized chemical synthesis. This study presents a synthetically accessible trillion-scale combinatorial chemical library constructed from a curated reaction set and over 1.8 million building blocks, providing a highly diverse and scalable resource for ultra-large virtual screening. In addition, we develop a Quadratic Unconstrained Binary Optimization (QUBO)-based strategy forresource-efficient compound synthesis, which maximizes building block utilization while generating molecules with desired properties. Together, this work establishes an integrated framework that bridges large-scale chemical space exploration with practical synthesis, enabling more efficient hit identification and downstream optimization in drug discovery.
Su et al. (Sat,) studied this question.