Increase Docking Score Screening Power by Simple Fusion With CNNscore.

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Structure-based virtual screening (SBVS) relies on scoring functions (SFs) to predict protein-ligand binding affinity.
• Traditional SFs have limitations due to approximations and simplifications.
• Deep learning (DL) based SFs show promise in capturing complex interactions.

Purpose of the Study:

• To develop and validate a novel docking-score fusion strategy integrating traditional and DL-based scores.
• To enhance the screening power of molecular docking for drug discovery.
• To identify novel inhibitors for the TYK2 target.

Main Methods:

• A fusion strategy combining pose scores from GNINA's convolutional neural network (CNN) with traditional docking scores was developed.
• The fusion strategy involved multiplying the Watvina docking score by the CNN score.
• The method was validated on diverse datasets and applied to screen nearly 12 billion molecules against TYK2.

Main Results:

• The fusion strategy demonstrated state-of-the-art screening power, outperforming traditional methods.
• Multiplying Watvina docking score by CNNscore proved effective.
• Two promising TYK2 inhibitors with IC50 values of 9.99 μM and 13.76 μM were identified in vitro.

Conclusions:

• The developed docking-score fusion strategy significantly enhances virtual screening capabilities.
• This integrated approach offers a powerful tool for identifying potent drug candidates.
• The method successfully identified novel inhibitors for the challenging TYK2 target.