LEADS-FRAG: A Benchmark Data Set for Assessment of Fragment Docking Performance

Area of Science:

• Computational chemistry
• Drug discovery
• Structural biology

Background:

• Fragment-based drug design (FBDD) is a key strategy in modern drug discovery.
• Molecular docking is a widely used computational method within FBDD.
• Assessing the accuracy of molecular docking programs for fragment placement is crucial.

Purpose of the Study:

• To introduce LEADS-FRAG, a novel benchmark dataset for evaluating fragment docking.
• To assess the fragment docking performance of popular small-molecule docking programs.
• To investigate the impact of rescoring on fragment docking accuracy.

Main Methods:

• Construction of the LEADS-FRAG dataset from 93 high-quality protein-fragment complexes from the Protein Data Bank.
• Evaluation of AutoDock, AutoDock Vina, FlexX, and GOLD using the LEADS-FRAG dataset.
• Analysis of docking pose accuracy using root mean square deviation (RMSD) and rescoring techniques.

Main Results:

• GOLD (with ChemPLP) and AutoDock Vina achieved the highest performance, generating near-native poses (>50%) in the top-ranked results.
• Across all poses, tested programs achieved up to 86% near-native conformations.
• Rescoring all poses with GOLD scoring functions and PLIF force field increased near-native conformation identification by up to 40%.

Conclusions:

• LEADS-FRAG provides a valuable resource for benchmarking fragment docking programs.
• Conventional small-molecule docking programs demonstrate satisfactory performance for fragment docking.
• Rescoring strategies significantly enhance the identification of near-native fragment poses.