Physics-Based Scoring

The physics-based scoring function is the recommended default for general-purpose docking in PandaDock. It provides a comprehensive molecular mechanics force field evaluation with excellent correlation to experimental binding affinities.

Overview

Scoring ID: physics_based

Type: Force field-based molecular mechanics scoring

Accuracy: R = 0.85 correlation with experimental binding affinities

Speed: 0.01-0.05 seconds per pose evaluation

Best for: General-purpose docking, balanced accuracy and speed, structural studies

Algorithm

The physics-based scoring function evaluates the following energy components:

\[E_{total} = E_{vdW} + E_{elec} + E_{desolv} + E_{hbond} + E_{torsion}\]

Energy Components

  1. Van der Waals (vdW) Energy

    \[\begin{split}E_{vdW} = \\sum_{i,j} 4\\epsilon_{ij} \\left[ \\left(\\frac{\\sigma_{ij}}{r_{ij}}\\right)^{12} - \\left(\\frac{\\sigma_{ij}}{r_{ij}}\\right)^6 \\right]\end{split}\]
    • Lennard-Jones 12-6 potential

    • Accounts for favorable dispersion interactions and steric clashes

    • Distance-dependent with r{v attractive and r{?? repulsive terms

  2. Electrostatic Energy

    \[\begin{split}E_{elec} = \\sum_{i,j} \\frac{q_i q_j}{4\\pi\\epsilon_0 \\epsilon_r r_{ij}}\end{split}\]
    • Coulombic interactions between partial atomic charges

    • Distance-dependent dielectric: ?(r) = 4r

    • Accounts for charge-charge, charge-dipole interactions

  3. Desolvation Energy

    \[\begin{split}E_{desolv} = \\sum_i \\Delta G_{solv,i} \\cdot SA_i\end{split}\]
    • Implicit solvation using atomic solvation parameters

    • Surface area-based desolvation penalties

    • Accounts for hydrophobic and hydrophilic contributions

  4. Hydrogen Bonding

    \[\begin{split}E_{hbond} = \\sum_{HB} E_{HB} \\cdot f(r) \\cdot f(\\theta) \\cdot f(\\phi)\end{split}\]
    • Explicit hydrogen bond detection

    • Geometry-dependent scoring (distance, angle, dihedral)

    • Donor-acceptor pair identification

    • Favorable contribution for well-formed H-bonds

  5. Torsional Penalty

    \[\begin{split}E_{torsion} = \\sum_{rot} W_{rot}\end{split}\]
    • Entropy penalty for each rotatable bond

    • Accounts for conformational entropy loss upon binding

    • Configurable weight per rotatable bond

Parameter Set

PandaDock uses optimized parameters derived from:

  • Atomic radii and well depths: AMBER ff14SB force field

  • Partial charges: AM1-BCC for ligands, AMBER for proteins

  • Solvation parameters: Optimized on PDBBind dataset

  • H-bond parameters: Geometry-dependent scoring functions

  • Torsional weights: Calibrated on experimental data

Usage

Basic Usage

pandadock dock -r protein.pdb -l ligand.sdf \\
               --scoring physics_based \\
               --center 10 20 30 --box 20 20 20

The physics-based scoring is the default, so you can omit --scoring:

pandadock dock -r protein.pdb -l ligand.sdf \\
               --center 10 20 30 --box 20 20 20

With Energy Decomposition

pandadock dock -r protein.pdb -l ligand.sdf \\
               --scoring physics_based \\
               --decompose-energy \\
               --center 10 20 30 --box 20 20 20

This generates detailed energy breakdowns:

  • Van der Waals contribution

  • Electrostatic contribution

  • Desolvation energy

  • Hydrogen bonding energy

  • Torsional penalty

Per-Residue Contributions

pandadock dock -r protein.pdb -l ligand.sdf \\
               --scoring physics_based \\
               --per-residue-decomposition \\
               --center 10 20 30 --box 20 20 20

Output includes interaction energy for each protein residue, useful for:

  • Identifying key binding residues

  • Hotspot analysis

  • Mutagenesis studies

  • Structure-activity relationships

Performance Characteristics

Accuracy Benchmarks

Tested on standard benchmark sets:

Dataset

Correlation (R)

RMSE (kcal/mol)

PDBBind Core

0.85

1.82

CASF-2016

0.83

1.95

Astex Diverse

0.81

2.11

Speed Benchmarks

  • Small ligand (<20 atoms): 0.01-0.02 seconds/pose

  • Medium ligand (20-40 atoms): 0.02-0.03 seconds/pose

  • Large ligand (>40 atoms): 0.03-0.05 seconds/pose

Screening throughput:

  • Single pose evaluation: 20-100 poses/second

  • Full docking (20 poses): 3-5 ligands/minute

Success Rate

  • RMSD < 2?: 90-95% (with enhanced_hierarchical algorithm)

  • Top pose RMSD < 2?: 75-85%

  • Correct binding mode identification: 85-90%

Strengths and Limitations

Strengths

 Physically Meaningful

Based on established molecular mechanics principles

 Interpretable

Energy components can be analyzed individually

 Broadly Applicable

Works well across diverse protein families and ligand types

 Balanced Accuracy

Good trade-off between speed and accuracy

 Per-Residue Analysis

Enables detailed interaction analysis

Limitations

 Solvation Approximation

Implicit solvation less accurate than explicit water models

 Fixed Charges

Doesn’t account for polarization or charge transfer

 No Entropy Terms

Only configurational entropy via torsional penalty

 Medium Speed

Slower than empirical scoring, faster than QM methods

Best Practices

Optimization Tips

For Maximum Accuracy:

pandadock dock -r protein.pdb -l ligand.sdf \\
               --scoring physics_based \\
               --rescoring mmgbsa \\
               --num-poses 50

For Better Speed:

pandadock dock -r protein.pdb -l ligand.sdf \\
               --scoring physics_based \\
               --fast \\
               --num-poses 10

For Detailed Analysis:

pandadock dock -r protein.pdb -l ligand.sdf \\
               --scoring physics_based \\
               --decompose-energy \\
               --per-residue-decomposition \\
               --visualize

Output Format

Energy Values

All energies reported in kcal/mol:

{
  "binding_affinity": -8.5,
  "energy_components": {
    "vdw_energy": -35.2,
    "electrostatic_energy": -12.4,
    "desolvation_energy": 28.1,
    "hbond_energy": -6.8,
    "torsional_penalty": 2.3
  }
}

Per-Residue Decomposition

{
  "residue_contributions": [
    {"residue": "TYR39", "energy": -2.8},
    {"residue": "ASP189", "energy": -3.5},
    {"residue": "LEU99", "energy": -1.2}
  ]
}

Comparison with Other Scoring Functions

Scoring

Accuracy

Speed

Analysis

physics_based

High

Medium

Detailed

empirical

Medium

Fast

Limited

precision_score

High

Slow

Very Detailed

hybrid

Very High

Slow

ML-enhanced

When to choose physics_based over alternatives:

  • vs empirical: Need better accuracy, willing to sacrifice speed

  • vs precision_score: Want faster results, don’t need extreme precision

  • vs hybrid: Don’t have GPU, want interpretable physics-based results

Examples

Basic Docking with Physics-Based Scoring

pandadock dock -r 1hsg_protein.pdb -l indinavir.sdf \\
               --center 15.0 20.0 25.0 \\
               --box 20 20 20 \\
               --num-poses 20 \\
               -o results_physics/

Virtual Screening with Physics-Based Scoring

pandadock dock -r kinase.pdb -l library_500.sdf \\
               --algorithm monte_carlo_cpu \\
               --scoring physics_based \\
               --fast \\
               --num-poses 5 \\
               -o screening_physics/

High-Accuracy Docking with Energy Analysis

pandadock dock -r protein.pdb -l ligand.sdf \\
               --algorithm enhanced_hierarchical_cpu \\
               --scoring physics_based \\
               --decompose-energy \\
               --per-residue-decomposition \\
               --rescoring mmgbsa \\
               --num-poses 50 \\
               --visualize \\
               -o detailed_analysis/

See Also