Molecular Dynamics Simulations

Validate binding, study dynamics, and calculate free energies with MD simulations

Molecular dynamics (MD) simulations model the motion of atoms over time, providing insights into protein flexibility, binding stability, and free energies that static methods cannot capture.

Overview

Use MD simulations to:

• Validate docking predictions
• Study protein conformational changes
• Calculate binding free energies
• Identify stable vs. unstable complexes
• Understand mutation effects
• Guide lead optimization

Simulation Types

Classical MD

Standard simulations with AMBER, CHARMM, or OPLS force fields.

Enhanced Sampling

Metadynamics, replica exchange to explore conformational space faster.

Free Energy

FEP, TI, MM-PBSA for quantitative binding affinity predictions.

QM/MM

Quantum mechanics for reactive sites combined with MM for environment.

Quick Start: Basic MD Simulation

1. Prepare System

   Upload protein-ligand complex from docking or experimental structure.

2. Setup Simulation

   Choose parameters:

   • Force field: AMBER ff14SB (recommended for proteins)
   • Water model: TIP3P (standard)
   • Box shape: Rectangular or octahedral
   • Box size: 10-12 Å padding around protein
3. Configure Protocol

   Standard protocol:

   • Minimization: 5000 steps
   • Heating: 0 → 300 K over 100 ps
   • Equilibration: 500 ps NPT
   • Production: 10-100 ns
4. Launch Simulation

   LiteFold runs on GPU-accelerated cloud compute. 100 ns typically completes in 4-8 hours.

5. Analyze Results

   Automated analysis includes:

   • RMSD/RMSF plots
   • Hydrogen bond analysis
   • Binding pocket volume
   • Interaction fingerprints

System Setup

Protein Preparation

LiteFold automatically:

• Adds missing atoms/residues
• Assigns protonation states (pH 7.4 default)
• Caps termini
• Adds counter-ions for neutralization
• Solvates in water box

Ligand Parameterization

For small molecules:

• GAFF2 (General AMBER Force Field) for drug-like molecules
• AM1-BCC charges (fast) or RESP (more accurate)
• Automatic tautomer/protonation state assignment

Complex Preparation

For protein-ligand complexes:

• Ligand positioned in binding site
• Waters kept or removed (configurable)
• Membrane proteins embedded in lipid bilayer
• Cofactors and ions included

Simulation Protocols

Short MD (10-50 ns)

Purpose: Quick validation, qualitative insights

Use for:

• Checking if docked pose is stable
• Preliminary comparison of compounds
• Identifying obvious instabilities

Limitations: May not sample rare events

Standard MD (50-200 ns)

Purpose: Reliable binding mode validation

Use for:

• Validating top docking hits
• Comparing binding stability across compounds
• Identifying key interactions
• Lead optimization guidance

Recommended for most drug discovery applications

Long MD (200-1000 ns)

Purpose: Conformational sampling, rare events

Use for:

• Flexible proteins and allosteric sites
• Induced fit mechanisms
• Slow conformational transitions
• Comprehensive dynamics studies

Note: More expensive, usually reserved for priority targets

Enhanced Sampling

Metadynamics: Add bias to explore conformational space faster

Replica Exchange: Run multiple simulations at different temperatures

Steered MD: Pull ligand out to study unbinding pathway

Use when: Standard MD insufficient for sampling timescales of interest

Analysis Tools

RMSD (Root Mean Square Deviation)

Measures structural drift from starting structure.

Interpretation:

• < 2 Å: Very stable
• 2-3 Å: Stable, small adjustments
• 3-5 Å: Moderate changes, check if converged
• > 5 Å: Large changes, may need longer simulation

RMSF (Root Mean Square Fluctuation)

Measures per-residue flexibility.

Interpretation:

• High RMSF: Flexible regions (loops, termini)
• Low RMSF: Rigid regions (core, binding site)
• Compare to B-factors from crystal structures

Hydrogen Bond Analysis

Tracks H-bonds over time:

• Identify persistent vs. transient interactions
• Key H-bonds present > 50% = important
• Loss of key H-bonds = unstable binding

Contact Analysis

Monitor protein-ligand contacts:

• Hydrophobic contacts
• Aromatic interactions
• Salt bridges
• Counts and distances tracked over time

Binding Pocket Volume

Track pocket size fluctuations:

• Is pocket breathing?
• Does ligand induce pocket opening/closing?
• Relevant for induced fit mechanisms

Free Energy Calculations

MM-PBSA/GBSA

Speed: Fast (post-processing of MD trajectory)

Accuracy: Semi-quantitative, good for ranking

Use for:

• Ranking analogs
• SAR understanding
• Identifying favorable modifications

Workflow:

1. Run MD simulation (50-100 ns)
2. Extract snapshots (every 10-50 ps)
3. Calculate energy for each snapshot
4. Average to get ΔG_bind

Free Energy Perturbation (FEP)

Speed: Slow (requires many simulations)

Accuracy: Quantitative (± 1 kcal/mol)

Use for:

• Precise affinity predictions
• Lead optimization decisions
• R-group optimization
• Prioritizing synthesis

Workflow:

1. Define perturbation (e.g., R = H → CH₃)
2. LiteFold creates alchemical pathway
3. Runs simulations for intermediate λ states
4. Calculates ΔΔG between compounds

Thermodynamic Integration (TI)

Similar to FEP but different mathematical approach. Often more robust for large perturbations.

GPU Acceleration

LiteFold uses GPU-accelerated MD engines:

• OpenMM: Flexible, supports custom force fields
• AMBER: Gold standard for biomolecular simulations
• GROMACS: High performance for large systems

Performance:

• 100 ns protein (50K atoms): 4-6 hours
• 100 ns protein-ligand (60K atoms): 5-8 hours
• 100 ns membrane protein (150K atoms): 12-18 hours

Best Practices

Run triplicates: For important compounds, run 3 independent simulations with different initial velocities to assess reproducibility.

Check equilibration: Ensure RMSD plateaus before analyzing. If trending upward, extend equilibration.

Compare to controls: Always simulate apo protein and known binders as references.

Start short, extend if needed: Begin with 50 ns. If not converged, extend to 100 or 200 ns.

Force field limitations:

• Cannot model bond breaking/forming
• May not accurately capture polarization
• Metal coordination requires special treatment
• Validate critical findings experimentally

Membrane Protein Simulations

Special setup for membrane proteins:

1. Membrane placement: Orient protein in lipid bilayer
2. Lipid selection: POPC, POPE, cholesterol mixtures
3. Equilibration: Longer protocol for lipid relaxation (1-2 ns)
4. Box size: Larger to accommodate membrane

LiteFold provides automated membrane embedding via Rosalind.

Cofactor and Metal Simulations

For proteins with cofactors:

• Metals: Use restraints or dummy bonds for coordination
• Cofactors: Parameterize separately (ATP, NAD, heme)
• Validation: Check coordination geometry throughout simulation

Mutation Studies

Compare wild-type vs. mutant:

1. Create mutant structure (in silico mutation)
2. Run MD for both WT and mutant
3. Compare:
   • Structural stability (RMSD)
   • Binding affinity (MM-PBSA)
   • Interaction patterns
   • Pocket geometry

Applications:

• Understand drug resistance mutations
• Predict mutation effects
• Design selectivity

Trajectory Visualization

Interactive viewer for MD trajectories:

• Play/pause/scrub through simulation
• Highlight changing interactions
• Create movies for presentations
• Export key frames

Example: Validating EGFR Inhibitor

Let’s validate a docked EGFR inhibitor with MD:

1. Prepare Complex

   • Protein: EGFR kinase domain (AlphaFold prediction)
   • Ligand: Docked erlotinib analog
   • Keep key water molecule in binding site
2. Setup Simulation

   • Force field: AMBER ff14SB + GAFF2
   • Water: TIP3P, 10 Å padding
   • Total atoms: 58,432
3. Run Protocol

   • Minimization + heating: 150 ps
   • Equilibration: 1 ns
   • Production: 100 ns (5 hours on GPU)
4. Analysis

   • RMSD (protein): 1.8 Å (stable)
   • RMSD (ligand): 1.2 Å (stable pose)
   • Key H-bonds:
     • Met793 (98% occupancy) ✓
     • Thr854 (87% occupancy) ✓
     • Asp855 (62% occupancy) ✓
   • MM-PBSA ΔG = -42 kcal/mol
5. Conclusion

   Binding mode is stable. Compound prioritized for synthesis.

Integration with Other Tools

MD simulations integrate with:

• Docking: Validate docking poses
• De Novo Design: Optimize generated molecules
• FEP: Rank analogs by affinity
• Rosalind AI: Ask “Why is this compound more stable?”

Next Steps

Binding Affinity

Calculate precise binding free energies with FEP

Drug Discovery Workflow

Integrate MD into complete discovery campaigns

De Novo Design

Use MD insights to design better molecules

Protein Analysis

Deep dive into protein dynamics and function