Abstract
Fragment-based drug discovery (FBDD) is a widely used strategy in early-stage drug development, but accurately predicting the binding affinities of fragments and their elaborated analogs poses unique computational challenges. These difficulties arise from weak binding affinities, diverse chemical scaffolds, and limited structural overlap between fragments and their optimized derivatives. While several free-energy methods exist, few are tailored to the specific requirements of FBDD. In this study, we evaluate the Separated Topologies (SepTop) approach for modeling fragment-based transformations, including fragment merging and linking. Using retrospective data sets from Cyclophilin D and SARS-CoV-2 Macrodomain 1, we demonstrate that SepTop can recover experimental binding affinities with good accuracy across both fragment and lead-like compounds. These results support SepTop's suitability for fragment optimization and highlight its potential to extend the reach of binding free-energy calculations into earlier stages of drug discovery.