In our latest paper published in the Journal of Chemical Information and Modeling (JCIM) by the American Chemical Society, we introduce two innovative methods for exploring ultra-large chemical libraries using accurate 3D quantum mechanics-based descriptors. 

The rapid expansion of ultra-large chemical libraries has revolutionized drug discovery, providing access to billions of compounds. However, this growth poses relevant challenges for traditional virtual screening (VS) methods. To address these limitations, synthon-based approaches have emerged as scalable alternatives, exploiting combinatorial chemistry principles to prioritize building blocks over enumerated molecules.  

With current library sizes, synthon-based approaches have 100k times lower computational cost than brute force methods. Since chemical spaces are growing rapidly and synthon-based methods scale better, this gap will continuously grow. 

In this work, we present exaScreen and exaDock, two novel synthon-based methodologies designed for ligand-based and structure-based VS, respectively.  

exaScreen: Our 3D ligand-based tool for virtual screening exploits atomic descriptors to select the optimal synthons through similarity measurements of the 3D molecular fields generated by the synthons and the reference fragment. For the set of reference compounds considered, the overall accuracy of exaScreen compares with the results obtained with brute force, as noted in the similar recovery of actives and the significant degree of identity between the actives selected by the two methods but at a significantly lower computational cost. exaScreen explores 70 billion compounds in less than 7 hours in a single workstation and requires less than 6 GB of disk storage. 

exaDock: Our new structure-based method for ultra-large libraries exploits a restrained docking to explore hybrid compounds (fragment reference + synthon) recurrently for the distinct reference fragments, which ultimately leads to the selection of the optimal synthon-based compounds built from the best ranked synthons identified for the reference fragments. The exaDock methodology can perform a search of EnamineREAL with a 50K lower computational cost than brute force. 

We evaluated both approaches across 11 diverse biological targets, including kinases and GPCRs. We show that exaScreen and exaDock achieve recovery rates comparable to brute-force screening strategies, offering computationally efficient strategies for VS in ultra-large chemical spaces using accurate 3D descriptors.  

One particularly interesting observation is the difference in scaffold overlap between our proposed methods and brute-force methods. In ligand-based, there is a significant overlap among the recovered active scaffolds while, in structure-based, the overlap is lower. This difference is explained by the influence of the geometrical and physicochemical properties of the binding site, particularly to the compactness of the pocket and to the solvent exposure of the reference fragments.