Abstract
In this paper, we present a dataset of homology-modeled structures of SARS-CoV-2 Spike protein Receptor Binding Domain (RBD) variants complexed with antigen-binding fragments (Fab) derived from the PDB structures 8GPY (Omicron BA.4/5 RBD in complex with a neutralizing antibody scFv), 8H7Z (BA.2 RBD in complex with BA7535 Fab), and 8XE9 (XBB.1.5 RBD in complex with BD55-1205). The dataset consists of six sets of complex structures generated by combining 14 RBD variants with three different Fab pairs. The SARS-CoV-2 Spike protein sequence variants included in the dataset are Wild Type, BETA, EPSILON, IOTA, ETA, GAMMA, LAMBDA, KAPPA, BA.2, BA.4, XBB.1.5, EG.5.1, and KP.2. Notably, the KP.2 variant, which has gained attention due to its inclusion in recent vaccine updates (https://www.ema.europa.eu/), is also part of this dataset. The models were refined using the Schrödinger Bioluminate suite of software. For each variant, the homology-modeled RBDs were complexed with the Fab fragments from 8GPY, 8H7Z, and 8XE9 by grafting them onto the RBDs, followed by energy minimization, both individually and in combination. This process resulted in a comprehensive dataset of RBD-Fab complexes, suitable for comparative analysis and further investigations. Binding affinities between the RBD-Fab pairs were calculated using the Prime MM-GBSA tool (VSGB solvation model and OPLS4 force field), enabling the ranking of antigen-antibody interactions. Structural minimizations were performed to accurately estimate interaction energies, providing insights into the efficacy of antibody binding across different variants. This dataset provides high-quality structural data that can be reused for in-depth studies of antibody-antigen interactions, particularly in the context of vaccine efficacy and viral immune evasion strategies. The inclusion of the KP.2 variant, which is central to the latest COVID-19 vaccine updates, makes these homology models especially relevant. Supplementary files include the homology-modeled structures, computed binding affinities, and quality assessments (QMEANDisCo) for each model, ensuring reproducibility and reliability for further analyses.