Normalizing molecular docking rankings using virtually generated decoys.

Area of Science:

• Computational Chemistry
• Drug Discovery
• Bioinformatics

Background:

• Virtual screening via molecular docking is crucial for identifying active compounds in drug discovery.
• Current docking scoring functions exhibit bias due to the physical properties of docked molecules, affecting accuracy.
• Large compound libraries inevitably contain molecules with diverse physical properties, necessitating bias correction.

Purpose of the Study:

• To present a method for normalizing docking scores to improve the accuracy of virtual screening.
• To address and account for the inherent bias in scoring functions caused by molecular physical properties.
• To introduce virtually generated decoy sets as a tool for evaluating and improving scoring functions.

Main Methods:

• Generation of property-matched decoy sets for each molecule in a screening library.
• Docking of library molecules and their corresponding decoys using a state-of-the-art docking method.
• Normalization of raw docking scores by comparing them against the scores of their matched decoys.

Main Results:

• Normalized scores represent the probability of a docking score originating from a non-active, property-matched decoy distribution.
• The method effectively accounts for biases introduced by the physical properties of screened molecules.
• Analysis of decoy sets provides insights into the reliability and limitations of scoring functions.

Conclusions:

• The proposed normalization method improves the reliability of hit identification in virtual screening.
• Property-matched decoy sets are valuable for assessing, refining, and developing more accurate docking scoring functions.
• This approach enhances the utility of molecular docking in accelerating drug discovery pipelines.