Complex structures, the solvent model, and minimization toggle for MM-GBSA calculations.

You provide a docked protein-ligand complex — most often straight from a DiffDock job — and can choose the implicit solvent model. The pose quality matters most: without MD sampling, MM-GBSA scores the input pose as given, so a clean, non-clashing complex is what makes the energy meaningful.

Webapp input methods

The .bloom* webapp provides several input methods for MM-GBSA calculations.

Input type selection

Input Type	Description	Example
From DiffDock	Docked complex from a previous DiffDock job — Ivy carries the pose's PDB text into `complex_pdb`	Select job from canvas or panel
Complex PDB (`complex_pdb`)	Single PDB file with protein and ligand	Upload `complex.pdb`
Protein + Ligand (`protein_pdb` + `ligand_sdf`)	Separate protein PDB and ligand SDF	Upload `protein.pdb` + `ligand.sdf`
From previous job	Complex from any docking tool output, carried into `complex_pdb` the same way	Select from completed jobs

There's no separate S3-URI parameter — this backend takes structure content directly, so a stored complex is read and its text passed as complex_pdb.

Structure requirements

Protein requirements

Standard amino acids: All 20 canonical residues fully supported
Non-standard residues: Modified amino acids (phosphoserine, methylated lysine, etc.) may cause parameterization failures; remove or replace with standard equivalents
Missing atoms: All heavy atoms in binding site residues must be present; missing atoms cause errors
Protonation: Histidine tautomers (HID, HIE, HIP) and cysteine protonation affect results significantly
Disulfide bonds: Detected automatically from PDB SSBOND records or distance-based criteria

Ligand requirements

Supported elements: C, H, N, O, S, P, F, Cl, Br, I
Charge states: Formal charges on the ligand must be correct for the binding site pH
Hydrogen atoms: Added automatically if missing; explicit hydrogens preferred for accuracy
Metal coordination: Metal-containing ligands require special handling and may not be fully supported

Complex requirements

Proximity: Ligand must be within the protein binding site (not solvent-exposed)
Single ligand: One ligand per calculation; multi-ligand complexes require separate runs
Chain selection: For multi-chain proteins, all chains in contact with the ligand are included
Water molecules: Crystallographic waters are removed by default (implicit solvent replaces them)

Force field

The tool always parameterizes with AMBER14 (ff14SB for the protein, GAFF2 for the ligand) — there is no force-field parameter to change it.

Implicit solvent models

solvent_model selects the Generalized Born term:

Value	Full Name	Speed	Accuracy	Description
`gb_obc2` (default)	Onufriev-Bashford-Case II	Fast	Good	Best general-purpose GB model; accurate for most systems
`gb_obc1`	Onufriev-Bashford-Case I	Fast	Slightly lower	Earlier OBC generation; kept for comparison
`gb_hct`	Hawkins-Cramer-Truhlar	Fastest	Approximate	Legacy option; lower accuracy for polar interactions

Choosing a solvent model

The "accuracy" column above is a general characterization of the GB models, not a measurement on this tool. Where it has been tested head to head, GB_OBC1 outperformed GB_OBC2 on four of six protein systems and was the best of the three models overall — see agreement with wet lab. Do not read gb_obc2 as the most accurate option simply because it is the default
Use gb_hct only for rapid screening when speed is more important than precision

Calculation parameters

These are the actual parameters this tool accepts:

Parameter	Type	Default	Description
complex_pdb	string	—	Protein-ligand complex PDB content. Supply this, or `protein_pdb` + `ligand_sdf`
protein_pdb	string	—	Protein PDB content (use with `ligand_sdf`)
ligand_sdf	string	—	Ligand SDF content, with 3D coordinates (use with `protein_pdb`)
solvent_model	string	`gb_obc2`	Implicit solvent model: `gb_obc2`, `gb_obc1`, or `gb_hct`
ionic_strength	float	0.15	Ionic strength in mol/L, range 0–1.0
minimize_complex	boolean	true	Run energy minimization before evaluation
decompose_residues	boolean	true	Inert. Per-residue decomposition is not implemented — there is no ΔG_bind to decompose. The backend does not declare this parameter, so it is dropped; the backend's own `include_per_residue` stays `false` and is rejected with an error if set true

Each call scores one complex. To score several DiffDock poses (top 1, top 5, or all), Ivy makes one mmgbsa_calculate call per pose — there's no poses selector on the tool itself. There's also no way to choose a force field, tune minimization steps/algorithm/tolerance, or set backbone restraints, surface tension, or a non-bonded cutoff — those aren't exposed parameters here.

Input preprocessing

Automatic preprocessing steps

When a complex is submitted, .bloom* performs:

Hydrogen addition: Missing hydrogens are added based on pH 7.4
Protonation assignment: Histidine tautomers assigned based on local hydrogen bonding
Water removal: Crystallographic waters removed (implicit solvent)
Ion removal: Counter-ions and buffer molecules removed
Ligand parameterization: GAFF2 parameters assigned automatically
Charge calculation: AM1-BCC charges computed for the ligand

Manual preprocessing recommendations

Protonation: If your binding site has unusual pH, manually set protonation states before submission
Missing loops: Rebuild missing protein loops before MM-GBSA (low-confidence ESMFold regions may affect results)
Alternate conformations: Remove alternate conformers; keep the A conformation
Multi-model PDB: Only the first model is used; split NMR ensembles if needed

Previous job integration

Compatible tools

DiffDock: Primary source of docked poses (most common workflow)
Chained docking runs: complexes from a structure → docking chain the orchestrator ran for you
Other docking tools: Any tool producing PDB-format complexes

Data handling

Pose selection: Choose specific poses from multi-pose docking results
Batch scoring: Score multiple complexes from a virtual screening campaign
Provenance tracking: Maintains connection to the original docking job

Batch processing strategies

Single complex scoring

Recommended for detailed analysis with per-residue decomposition
Use all available poses from DiffDock for comprehensive ranking

Compound series scoring

Score one pose per compound against the same target
Use consistent force field and solvent model across all compounds
Compare relative binding energies rather than absolute values

Virtual screening rescoring

Score top hits from a docking campaign
Use gb_obc2 for the best accuracy-speed tradeoff, or gb_hct when speed matters more than precision
Filter by DeltaG_bind threshold before detailed analysis

MM-GBSA inputs and parameters