Interpreting MM-GBSA outputs

MM-GBSA returns single-point energies for the minimized protein-ligand complex, together with a caveat explaining what they are not. This implementation does not compute a binding free energy: delta_g_bind is null and binding_affinity_available is false on every run, because the job evaluates the complex as one system rather than decomposing it into receptor and ligand. Read how MM-GBSA results compare to the bench before you act on anything here.

Output artifacts

Each completed MM-GBSA job returns single-point energies for the minimized complex, plus a method and caveat field. It does not return a binding free energy, an energy decomposition, or per-residue analysis — see the field table below for what is and is not emitted.

Primary output fields

There is no binding strength classification

Earlier versions of this page described a mapping from DeltaG_bind to "very strong / strong / moderate / weak / no binding" with an estimated Kd in nanomolar. That mapping has been removed, from the backend and from these docs, and it should not be reintroduced. Two independent reasons:

• It had no valid input. The classifier was being fed the minimized complex's raw total potential energy — thousands of kcal/mol — mislabelled as a binding free energy, so it returned "very strong / drug-like" for every run. Nothing in the running code now emits a classification or a Kd.
• The conversion would not be defensible even with a real ΔG. Kd = exp(ΔG/RT) amplifies the input error geometrically: 1 kcal/mol of error is a 5.4-fold error in Kd, 2 kcal/mol is 29-fold, 3 kcal/mol is 158-fold. Published MM/GBSA absolute errors run from 4.4 to 21 kcal/mol depending on the target. The full argument, with the primary citations, is in how MM-GBSA results compare to the bench.

What the method is good for, where it produces a ΔG at all, is ranking a congeneric series against one target — not stating an affinity. Report the energy with its units and keep the error bars.

Energy decomposition

Understanding each term

Van der Waals (DeltaG_vdw):

• Represents shape complementarity between protein and ligand
• Typically the largest favorable contribution for hydrophobic ligands
• More negative values indicate better shape fit in the binding pocket

Electrostatic (DeltaG_elec):

• Represents charge-charge and dipole interactions
• Favorable for ligands with hydrogen bonds or salt bridges to the protein
• Can be very favorable but is often offset by the polar solvation penalty

Polar solvation (DeltaG_polar):

• Represents the desolvation cost of burying polar groups upon binding
• Typically positive (unfavorable) and partially cancels the electrostatic term
• Large positive values indicate significant desolvation penalty

Non-polar solvation (DeltaG_nonpolar):

• Represents the hydrophobic effect of burying non-polar surface area
• Typically a small negative (favorable) contribution
• Proportional to the buried solvent-accessible surface area

Reading the decomposition chart

The canvas card displays a stacked bar chart with:

• Green bars for favorable (negative) energy terms
• Red bars for unfavorable (positive) energy terms
• Total DeltaG_bind shown as a separate summary bar
• Hover over each segment for exact values

Per-residue decomposition

When per-residue decomposition is enabled, the output includes a ranked table of residues:

Interpreting per-residue hotspots

• Negative total: Residue contributes favorably to binding (hotspot)
• Positive total: Residue destabilizes the complex (potential site for improvement)
• Large magnitude: Residue has a significant impact on binding; mutations here will likely affect affinity
• Top 5-10 residues: Usually account for the majority of binding energy

Common interaction patterns

• Hydrophobic contacts: Large negative vdW term with near-zero electrostatic (e.g., Leu, Ile, Val, Phe)
• Hydrogen bonds: Significant negative electrostatic term, partially offset by positive polar solvation
• Salt bridges: Very large negative electrostatic term from Asp/Glu-ligand or Arg/Lys-ligand interactions
• Pi-stacking: Moderate negative vdW contribution from Phe, Tyr, Trp, or His

Multi-pose results

When multiple poses are scored, the output includes:

• Pose ranking table: All poses sorted by DeltaG_bind
• Best pose highlight: The top-scoring pose is marked and displayed in the 3D viewer
• Energy comparison: Side-by-side energy decomposition across poses
• Pose RMSD: Heavy-atom RMSD between poses for structural comparison

Downloading artifacts

• CSV export: Full energy decomposition, per-residue data, and pose rankings
• JSON export: Structured results with metadata and calculation parameters
• PDB files: Minimized complex structure for each scored pose
• PyMOL script: Visualization script with hotspot residues highlighted

Canvas visualization

When an MM-GBSA job completes, a result card appears on the canvas.

• Card size: 500x450px
• Border color: Purple (#9C27B0)
• Visualization: Binding energy chart with per-residue decomposition

Tier 2 metrics

• Best DG (kcal/mol)
• Ligand count
• Method type
• Mean DG
• Frame count

Interpreting canvas metrics

There is no threshold table here, because there is no binding free energy to threshold. The canvas card shows the energy values the job returned and a colour band on their magnitude; the band is a display convenience, not a binding verdict. A run with no affinity verdict renders the caveat instead.

Quality checks

• Check that binding_affinity_available is present and read the caveat string before quoting any number
• Confirm minimization converged (convergence_achieved) and look at minimization_steps
• Treat unusually large energy magnitudes as a parameterization signal, not as tight binding
• Inspect the upstream pose. A physics-based energy computed on a wrong pose is a precise answer to the wrong question, and nothing in the output flags it — see what the figure doesn't cover

Read next