Dataset for evaluating accuracy of Boltz-2's structure and binding affinity predictions

This dataset is used to generate all results and figures reported in the manuscript:

“On the Reliability of AI Methods in Drug Discovery: Evaluation of Boltz-2 for Structure and Binding Affinity Prediction”

Below are descriptions of individual files:

3CLPro_dataset.csv: 16,780 compounds for 3CLPro, consisting of SMILES strings, Boltz-2 predicted binding affinities and binding probability, and binding free energy from ESMACS calculations.

3CLPro-top100.csv: top 100 compounds based on Boltz-2 predicted binding affinities

3CLPro-top100-docked-structure.tar: compound structures from docking approach

3CLPro-top100-Boltz2-structure.tar: strcutures of protein-compound complexes from Boltz-2 prediction

3CLPro-top100-energies.csv: binding free energies from fine-grained ESMACS approach, initiiated from different structures

TNKS2_dataset.csv: 21,702 compounds for TNKS2, consisting of SMILES strings, Boltz-2 predicted binding affinities and binding probability, and binding free energy from ESMACS calculations.

TNKS2-top100.csv: top 100 compounds based on Boltz-2 predicted binding affinities

TNKS2-top100-docked-structure.tar: compound structures from docking approach

TNKS2-top100-Boltz2-structure.tar: strcutures of protein-compound complexes from Boltz-2 prediction

TNKS2-top100-energies.csv: binding free energies from fine-grained ESMACS approach, initiiated from different structures