Benchmarking HelixFold3-Predicted Holo Structures for Relative Free Energy Perturbation Calculations

Area of Science:

• Computational chemistry
• Structural biology
• Drug discovery

Background:

• Free energy perturbation (FEP) calculations are vital for predicting binding affinities in drug discovery.
• Accurate protein-ligand complex structures are crucial for FEP accuracy.
• AlphaFold2, while advanced, cannot predict holo structures, limiting its use in structure-based drug design.

Purpose of the Study:

• To evaluate HelixFold3's performance in predicting protein-ligand complexes for FEP calculations.
• To compare HelixFold3's accuracy against existing methods like AlphaFold2.
• To assess the utility of HelixFold3-predicted structures in early-stage drug discovery.

Main Methods:

• HelixFold3 was used to predict protein-ligand complexes for eight targets from a benchmark set.
• Binding site conformations were analyzed for accuracy.
• Free energy perturbation calculations were performed using predicted holo and apo structures, as well as crystal structures.

Main Results:

• HelixFold3 demonstrated superior performance in predicting binding site conformations compared to AlphaFold2.
• Predicted holo structures showed higher binding site accuracy than apo structures.
• FEP calculations using HelixFold3-predicted structures achieved accuracy comparable to crystal structures.
• HelixFold3 successfully predicted structures for novel ligands, maintaining FEP accuracy.

Conclusions:

• HelixFold3 accurately predicts protein-ligand complex structures, including holo conformations.
• HelixFold3-predicted structures can serve as effective substitutes for crystal structures in FEP calculations.
• This advancement supports the application of structure-based drug design in early drug discovery.