AutoShim: empirically corrected scoring functions for quantitative docking with a crystal structure and IC50 training...

Area of Science:

• Computational chemistry
• Drug discovery
• Structural biology

Background:

• Developing accurate scoring functions for high-throughput docking is challenging.
• Existing methods struggle to correlate with measured binding affinity.

Purpose of the Study:

• To present AutoShim, a novel method for improving binding affinity predictions in drug discovery.
• To adapt general scoring functions for specific protein targets using pharmacophore shims.

Main Methods:

• AutoShim utilizes the Magnet program to add pharmacophore-like shims to protein binding sites.
• Partial least-squares (PLS) regression weights the shims to optimize scoring functions against IC50 data.
• An iterative procedure refines poses and parameters, ensuring robust model convergence.

Main Results:

• AutoShim significantly improves affinity prediction accuracy for held-out compounds.
• The method demonstrates reproducible convergence in 2-4 iterations, avoiding overtraining.
• Simpler pharmacophore shims provide reasonable predictions and clear molecular interaction insights.

Conclusions:

• AutoShim offers a practical and effective approach to enhance binding affinity prediction in drug discovery.
• The method provides reproducible and interpretable results, aiding in the design of new drug compounds.
• This automated procedure requires a training set but adds minimal computational overhead.