Structure-Based Prediction of SARS-CoV-2 Variant Properties Using Machine Learning on Mutational Neighborhoods

This dataset presents a structure- and function-annotated resource of SARS-CoV-2 spike receptor-binding domain (RBD) variants. It includes 3,705 theoretical 1-step RBD mutations and 200 higher-order Omicron BA.1 and BA.2 variants, each predicted using AlphaFold2 and ESMFold and enriched with sequence-derived Bio2Byte descriptors. The dataset comprises six CSV files containing variant-level structural and empirical features, including RMSD, TM-score, plDDT, SASA, electrostatic potential, aggregation propensity, disorder, and flexibility, as well as ACE2 binding and RBD expression scores. The resource supports structure–function analysis, mutation impact prediction, and machine learning–based variant modeling for virology and pandemic preparedness.