Assessing How Residual Errors of Scoring Functions Correlate to Ligand Structural Features

Area of Science:

• Computational Chemistry
• Drug Discovery
• Molecular Modeling

Background:

• Scoring functions (SFs) are crucial for early-stage drug discovery but have moderate accuracy.
• Existing SFs often struggle with systematic improvement, despite target-specific successes.
• Ligand features, like hydrogen bonds and hydrophobic interactions, may contribute to SF errors.

Purpose of the Study:

• To investigate if specific ligand interaction features correlate with scoring function errors.
• To identify potential directions for systematic improvement of general scoring functions.
• To analyze residual errors of various SFs against experimental data.

Main Methods:

• Utilized the CASF-2016 dataset of 285 ligands.
• Calculated scores using multiple SFs (AutoDock 4.2, Vina, X-Score, NNScore2.0, ΔVina RF20, DSX).
• Correlated SF residual errors with the presence of specific ligand fragments and medicinal chemistry interactions.

Main Results:

• Scoring functions generally perform adequately but show room for improvement.
• Analysis revealed correlations between SF errors and specific ligand fragments.
• Better parameterization of fragment interactions can enhance SF accuracy.

Conclusions:

• The proposed method offers a pathway for systematic SF improvement without significant complication.
• Refining the handling of specific ligand-interaction fragments can boost SF performance.
• Data scarcity for ligand-receptor complexes currently limits the direct application of this approach.