Collinearity of Decomposed Energy Terms in MM-GBSA Binding Free Energy Calculations

Technical Summary: Collinearity of Decomposed Energy Terms in MM-GBSA Binding Free Energy Calculations

Problem Statement
The Molecular Mechanics-Generalized Born Surface Area (MM-GBSA) method is a standard end-state approach for estimating binding free energies in computational drug design. A common practice involves decomposing the total free energy into distinct components—van der Waals ($E_{vdW}$), electrostatic ($E_{elec}$), polar solvation ($G_{GB}$), and nonpolar solvation ($G_{SA}$)—to either correlate these terms with experimental data or assign specific physical meanings to them. This study challenges the fundamental assumption that these decomposed terms possess independent fitting coefficients. The authors posit that treating these terms as statistically independent predictors in regression models may be invalid due to inherent mathematical and physical redundancies within the MM-GBSA formalism.

Methodology
The authors employed a dual approach combining analytic derivations and large-scale molecular dynamics (MD) simulations to investigate the statistical relationships between energy terms.

• Analytic Derivation: The study mathematically examined the structural dependencies between the energy components, specifically focusing on how the Generalized Born (GB) model is designed to screen vacuum electrostatics.
• Simulation Data: The analysis utilized extensive MD trajectory data. The authors evaluated both instantaneous fluctuations within trajectories and averaged values across a diverse dataset of 139 protein-protein complexes.
• Protocol Comparison: The robustness of the findings was tested across different MM-GBSA protocols, including both single-trajectory and three-trajectory approaches.
• Entropy Analysis: The study also incorporated Interaction Entropy and $C_2$ entropy corrections to assess their dependence on underlying electrostatic fluctuations.

Key Contributions and Results
The primary contribution of this work is the systematic diagnosis of severe collinearity among the standard decomposed energy terms in MM-GBSA. The results demonstrate:

1. Electrostatic and Polar Solvation Redundancy: The protein-ligand Coulomb interaction energy ($E_{elec}$) and the GB solvation correction ($G_{GB}$) exhibit near-perfect collinearity ($R^2 \geq 0.99$). This reflects their designed relationship, where the GB term serves as a solvent screening mechanism for vacuum electrostatics.
2. Van der Waals and Nonpolar Solvation Correlation: Similarly, the van der Waals interaction energy and the nonpolar solvation term ($G_{SA}$) show strong correlation, as both are primarily driven by the buried surface area.
3. Entropy Dependence: Interaction entropy and $C_2$ entropy corrections were found to be strongly dependent on underlying electrostatic fluctuations, further reinforcing the redundancy within the energy decomposition.
4. Universality: These correlations persist regardless of whether the data is analyzed at the level of instantaneous trajectory fluctuations or as ensemble averages. The findings hold true across the diverse set of 139 protein-protein complexes and remain consistent in both single- and three-trajectory protocols.

Significance and Claims
The paper claims that the assumption of independent fitting coefficients for decomposed MM-GBSA terms is statistically unsound. Consequently, the authors caution against utilizing these individual terms as independent predictors in regression models, as doing so may lead to overfitting or misinterpretation of physical contributions.

Instead, the study suggests a more robust modeling strategy: combining the highly correlated terms into effective, aggregated contributions. Specifically, the authors propose grouping terms into effective polar, nonpolar, and entropic contributions. By acknowledging and addressing this collinearity, the study aims to facilitate more interpretable and statistically robust predictive modeling in binding free energy calculations.