FastBindRank dataset: Boltz-2 predictions of binding probability and log10(IC50) for 1.85 million random PubChem compounds targeting HDAC11

This dataset contains structure-based predictions generated by the Boltz-2 model for a randomly sampled subset of 1.85 million compounds from the 122-million-compound PubChem library (https://pubchem.ncbi.nlm.nih.gov/). The predictions include binding probabilities and estimated log10(IC50) values for compounds targeting histone deacetylase 11 (HDAC11), a class IV histone deacetylase with limited availability of selective inhibitors. This dataset was generated as part of the FastBindRank framework, which applies knowledge distillation to transfer high-fidelity structure-based predictions from Boltz-2 to a lightweight surrogate model (BindRankNet) for scalable virtual screening across ultra-large chemical spaces.