Pose Rescorer v1.1
Physics-based post-docking rescoring using MM/GBSA
Overview
Rescore is a command-line tool for post-docking rescoring of protein-ligand complexes using single-frame MM/GBSA calculations. It provides automated parameterization, complex assembly, and energy calculations through a streamlined interface to AmberTools.
Purpose: Rescore performs relative ranking of docked poses or compound series. It does not compute thermodynamically rigorous binding free energies.
Scope
This tool is designed for:
- Ranking docked ligand poses
- Prioritizing compounds from virtual screening
- Comparing binding modes qualitatively
- High-throughput rescoring of multiple ligands against one receptor
This tool is not designed for:
- Computing absolute binding free energies (ΔG)
- Predicting IC50 or Ki values
- Replacing rigorous alchemical free energy methods
- Publication-quality thermodynamic predictions
Scientific Positioning
Rescore performs single-frame MM/GBSA calculations—a fast, approximate method for estimating protein-ligand binding energies. The approach combines molecular mechanics force fields with implicit solvent models.
What It Does
- Parameterizes proteins with AMBER ff14SB force field
- Parameterizes ligands with GAFF2 force field and AM1-BCC charges
- Performs restrained energy minimization to remove docking clashes
- Computes MM/GBSA binding energies using generalized Born implicit solvent
- Reports ΔGbind from MM/GBSA without entropy (−TΔS omitted)
What It Does Not Do
- Molecular dynamics simulations
- Conformational sampling
- Entropy calculations (−TΔS = 0 in this implementation)
- Explicit solvent modeling
- Metal cofactor parameterization
Methodology
Force Fields
- Protein: AMBER ff14SB
- Ligand: GAFF2 (General Amber Force Field 2)
- Charges: AM1-BCC (semi-empirical quantum mechanics)
Solvation Model
- Method: Generalized Born (GB, igb=5 OBC II)
- Salt concentration: 0.15 M (physiological)
- Note: Poisson-Boltzmann (PB) supported but not recommended for single-frame rescoring
Minimization Protocol
By default, Rescore performs restrained energy minimization before scoring to relax steric clashes introduced by docking:
- Cycles: 100 (50 steepest descent + 50 conjugate gradient)
- Restraints: Protein backbone atoms (N, CA, C) restrained at 2.0 kcal/mol/Ų
- Free atoms: Ligand and protein side chains
- Solvation: GB implicit solvent with 12 Å cutoff
Important: The minimization step is not molecular dynamics. It is a local energy minimization to prepare structures for scoring. The calculation remains single-frame.
Use Cases
Appropriate
- Post-processing docking results for improved pose ranking
- Filtering large virtual screening libraries
- Comparing binding modes from different docking programs
- Preliminary compound prioritization for synthesis or testing
- Uncertainty quantification via Rapid Perturbation Sampling (RPS)
Inappropriate
- Reporting ΔG values as experimental predictions
- Correlating scores with experimental binding affinities quantitatively
- Using scores as training data for machine learning without validation
- Replacing rigorous free energy calculations (FEP, TI, metadynamics)
Optional Features
Plotting
Pose-Rescorer optionally generates bar plots summarizing single-frame MM/GBSA or MM/PBSA scores to provide a visual aid for result inspection; these plots are descriptive only and do not alter score interpretation.
Rapid Perturbation Sampling (RPS)
RPS is an optional diagnostic feature that quantifies numerical sensitivity of single-frame scores to small coordinate perturbations. It generates N perturbed copies of a ligand (via Gaussian noise, default σ = 0.2 Å), rescores each, and reports statistics (mean, std, CV, 95% CI).
Critical: RPS is not molecular dynamics, conformational sampling, or thermodynamic ensemble averaging. It is numerical sensitivity analysis for uncertainty quantification only.
Philosophy
Rescore is built on the principle that computational tools should be honest about their limitations. MM/GBSA is a fast, approximate method—not a replacement for thermodynamic rigor.
This tool provides relative energies for ranking purposes. The absolute values have large uncertainties and should not be interpreted as quantitative predictions of binding affinity.
For publication-quality binding free energy predictions, consider alchemical methods (FEP, TI) or enhanced sampling techniques with explicit solvent.
Recommended reading: Genheden & Ryde (2015) "The MM/PBSA and MM/GBSA methods to estimate ligand-binding affinities." Expert Opin. Drug Discov. 10(5):449-461.