Cross-docking of inhibitors into CDK2 structures. 2

Area of Science:

• Computational chemistry
• Drug discovery
• Structural biology

Background:

• Previous work evaluated Gold and Glide docking accuracy using single protein conformations for 150 CDK2/inhibitor complexes.
• Both programs showed high docking accuracy, with Glide exhibiting modest correlations between scoring and affinity (log IC50).

Purpose of the Study:

• To investigate the impact of consensus docking using multiple protein conformations on accuracy and affinity prediction.
• To assess the statistical significance and predictivity of affinity/scoring correlations.

Main Methods:

• Docking simulations were performed using multiple protein conformations in a consensus approach.
• Analysis focused on scores from correctly docked poses to evaluate log(IC50)/score correlations.
• Validation sets and scrambling experiments were employed to assess statistical significance.

Main Results:

• Consensus docking improved Glide's docking accuracy but not Gold's.
• Using multiple protein structures enhanced log(IC50)/score correlations when considering only correct poses.
• However, these improvements were attributed to overfitting artifacts rather than genuine scoring accuracy enhancement.

Conclusions:

• Docking to multiple protein conformations can lead to overfitting, creating an illusion of improved scoring accuracy.
• The study highlights the importance of rigorous validation to distinguish true predictive power from artifacts in computational drug discovery.