MM/PBSA binding free energy, explained
How end-state free-energy methods work, how to run them on an MD trajectory and how far to trust the numbers.
What MM/PBSA calculates
Molecular Mechanics Poisson–Boltzmann Surface Area (MM/PBSA) estimates the binding free energy of a ligand to a protein by averaging energy terms over snapshots from a molecular dynamics simulation. MM/GBSA is a faster variant that treats solvation with the Generalised Born model.
The energy terms
- Van der Waals and electrostatic interaction energies in vacuum.
- Polar solvation energy, from the Poisson–Boltzmann or Generalised Born model.
- Non-polar solvation energy, from the solvent-accessible surface area.
- Optionally an entropy term, which is expensive and noisy and is often left out when only ranking ligands.
How to run it
- Run a well-equilibrated production simulation of the complex.
- Choose a stable window of the trajectory using RMSD.
- Extract frames at regular intervals.
- Calculate the energy terms for the complex, receptor and ligand for each frame.
- Average the binding energy and report the standard deviation and error.
- Decompose the energy per residue to find the key contributors.
Where it goes wrong
- Using frames before equilibration or after the ligand has drifted.
- Comparing very different ligands, such as charged and neutral, with one setup.
- Reading absolute values as experimental affinity.
- Ignoring the spread between frames.
Related reading
Run the simulation first with the GROMACS workflow, and check stability with MD analysis metrics.
Frequently asked questions
MM/PBSA or MM/GBSA, which is better?
PB is more rigorous but slower, GB is faster. For ranking related ligands both are commonly used. Consistency across the series matters most.
Should entropy be included?
It is often omitted for ranking because it is costly and noisy. When omitted, the result is better called a binding energy estimate.
How many frames are needed?
Often 100 to 1000 frames from a stable window, depending on the length of the run and the variability of the system.
Can MM/PBSA predict IC50?
Not directly. It correlates with experimental affinity for similar ligands at best, and should be used as supporting evidence.
Related guides
Look for convergence, not single numbers. Combine several metrics before concluding that a ligand is stably bound.
GROMACS tutorialGROMACS Protein–Ligand MD Simulation WorkflowBuild topologies, solvate and neutralise, minimise, equilibrate in NVT then NPT, run production, then correct periodic boundaries before analysis.
Related service
Written by Sohel Akhtar Ansari, ReactorMD founder (ORCID). Computational predictions are not experimental confirmation.