DLIGAND2

DLIGAND2 predicts protein-ligand binding affinities using a knowledge-based energy function derived from the distance-scaled, finite, ideal-gas reference state (DFIRE) to improve accuracy in molecular docking.

Key Features:

• Expanded Atom Representation: Uses 167 residue-specific atom types instead of the previous 13 mol2 atom types for more detailed modeling of protein-ligand interactions.
• Updated Training Dataset: Employs an updated dataset comprising 12,450 monomer protein chains to enrich the statistical potential.
• Improved Predictive Performance: Delivers improved predictions for native complex structures and docking-generated poses and achieves a 52% increase in enrichment factors within the top 1% when evaluated against the DUD-E decoy set.
• Benchmark Superiority: Outperforms scoring functions such as Autodock Vina and exceeds empirical and machine-learning approaches across three benchmark tests and in virtual screening on targets not homologous to the DUD-E training set.
• Versatility in Applications: Functions as a parameter-free statistical potential for reassessing docking poses or as a component within other scoring functions in virtual screening and drug discovery workflows.

Scientific Applications:

• Structure-based molecular docking: Predicts and ranks binding affinities to evaluate and re-score docking poses.
• Virtual screening: Enriches top-ranked hits in screening campaigns, demonstrated using the DUD-E decoy set.
• Drug discovery and lead optimization: Supports medicinal chemistry and computational biochemistry efforts by providing affinity assessments for candidate ligands.
• Structural biology: Analyzes protein-ligand interaction patterns using residue-specific atom-level potentials.

Methodology:

Computes a knowledge-based statistical potential using the DFIRE reference state and 167 residue-specific atom types derived from an updated dataset of known protein structures (12,450 monomer chains) to calculate interaction energies for binding affinity prediction.