Structure-based prediction of SARS-CoV-2 variant properties using machine learning on mutational neighborhoods.

Abstract

This dataset presents a structure-enriched resource of theoretical and empirical SARS-CoV-2 spike receptor-binding domain (RBD) variants, developed under the STAYAHEAD project for pandemic preparedness. It integrates large-scale in silico structure predictions with empirical biophysical measurements. The dataset includes 3,705 single-point Wuhan-Hu-1 RBD variants and 100 higher-order Omicron BA.1/BA.2 variants, annotated with AlphaFold2 and ESMFold metrics and Bio2Byte sequence-based predictors. Structural descriptors-RMSD, TM-score, plDDT, solvent accessibility, hydrophobicity, aggregation propensity-are linked to ACE2 binding and expression data from deep mutational scanning. Provided as a FAIR² Data Package, it supports structure-function analysis, variant modeling, and responsible reuse in virology, structural biology, and computational protein science. This collaboration was co-funded by the PPP Allowance from Health ∼ Holland, Top Sector Life Sciences and Health, to stimulate public-private partnerships.