OpenFE
v1.13.0Absolute hydration free energy calculations for neutral small molecules
The OpenFE webserver runs the absolute hydration free energy (AHFE) protocol of OpenFE 1.13, the open-source alchemical free energy toolkit from the Open Free Energy consortium. You submit one neutral small molecule as an SDF file with explicit hydrogens and 3D coordinates, and the server returns its hydration free energy in kcal/mol, the spread across repeats, per-leg estimates, convergence diagnostics, and every file the OpenFE run writes. Simulations run on GPU with OpenMM.
The hydration free energy is the free energy change when a molecule moves from vacuum into water. Negative values mean the molecule prefers water. It is used to check force field quality for a chemical series, as a solvation term in solubility and permeability work, and as a fast test of a free energy setup before a more expensive binding calculation.
A free energy is a state function, so the path between two states does not have to be physical. Instead of moving the molecule from gas to water, OpenFE switches its interactions off in two separate environments and closes the thermodynamic cycle:
A coupling parameter moves each leg through intermediate states (14 in the solvent leg and 5 in the vacuum leg, because a molecule alone has no solvent interactions to switch off gradually), and the states exchange configurations by Hamiltonian replica exchange. Free energy differences along each leg come from MBAR (multistate Bennett acceptance ratio), which uses the energies of every sample in every state. OpenFE reports
averaged over the repeats.
Each repeat of each leg parameterizes the system (OpenFF 2.2.1 small-molecule force field with AM1-BCC charges from AmberTools, and OpenMM's Amber TIP3P water and ion files by default), runs a short conventional MD equilibration, builds the alchemical system, and then runs multistate equilibration and production at 298.15 K and 1 bar. These protocol settings are OpenFE's defaults and are not changed by the server.
OpenFE is billed for GPU runtime at 20 credits per minute. The quote shown before submission is an estimate from the simulation length and number of repeats:
| Simulation length | 1 repeat | 3 repeats |
|---|---|---|
Test | 100 | 300 |
Short | 500 | 1,500 |
Medium | 1,000 | 3,000 |
Production | 2,800 | 8,400 |
The default calculation, Production with 3 repeats, is quoted at 8,400 credits. Credits are reserved in steps while the run continues and actual runtime is settled at the end, so a faster run costs less than its quote. A run stops at eight hours (9,600 credits) or when the workspace runs out of credits; in that case no hydration free energy is reported and the files written so far are returned. The water model does not change the quote. Molecule size changes runtime but not the quote.
| Input | Accepted value |
|---|---|
Ligand | One neutral small molecule as an .sdf file up to 10 MB, with explicit hydrogens and 3D coordinates. |
The SDF is used exactly as submitted: its coordinates, atoms and hydrogens go to OpenFE unchanged. OpenFE simulates only the atoms it receives, so an SDF without hydrogens is simulated without them, and the submitted conformer is the starting geometry and the geometry used for AM1-BCC charges. SMILES is not accepted because OpenFE has no SMILES input for its simulation protocols; convert it first with SMILES to SDF with Add hydrogens on, or prepare the structure with Open Babel, and check that hydrogens are present.
The protocol accepts one molecule per job. A multi-record SDF, a charged molecule and a molecule containing a metal are rejected before submission, and an SDF without coordinates is rejected when the run starts. OpenFE's absolute solvation protocol does not support charged molecules, and its default AM1-BCC charges and OpenFF force field do not cover metal-containing or coordination-complex ligands.
| Parameter | Type | Default | Description |
|---|---|---|---|
Simulation Length | enum | Production | Multistate production sampling per leg: Test 10 ps, Short 0.5 ns, Medium 2 ns solvent and 1 ns vacuum, Production 10 ns solvent and 2 ns vacuum. |
Number of Repeats | enum | 3 | Independent repeats of both legs: 1 or 3. Three is OpenFE's default. |
Production matches OpenFE's default protocol. The other lengths shorten only the multistate production phase, and Test also shortens multistate equilibration to 10 ps. Conventional MD equilibration runs at OpenFE's default length in every profile, so even Test includes 0.8 ns of solvent MD before the alchemical simulation. Test checks that a molecule parameterizes and runs; it is not long enough for a scientific estimate.
| Parameter | Type | Default | Description |
|---|---|---|---|
Water model | enum | TIP3P | TIP3P or TIP4P-Ew. TIP4P-Ew uses its matching OpenMM water and ion files in both legs. |
The water model is part of the force field, so the same molecule can give a different hydration free energy with each model.
The result table has one row for the submitted molecule:
| Column | Meaning |
|---|---|
Ligand (SMILES) | Canonical SMILES of the simulated molecule. |
ΔG (kcal/mol) | Hydration free energy, vacuum leg minus solvent leg, averaged over repeats. |
± (kcal/mol) | Standard deviation of the repeat estimates of both legs, combined in quadrature. |
Minimum MBAR overlap | Smallest MBAR overlap scalar across all legs and repeats. |
Forward/reverse complete | Whether OpenFE's forward and reverse convergence analysis finished for every repeat. |
Simulation | Selected simulation length. |
Repeats | Number of repeats. |
Water model | Selected water model. |
Time (s) | Wall time of the OpenFE calculation. |
The Files view returns these documents together with every file the run writes:
| File | Contents |
|---|---|
openfe_ahfe_results.json | The estimate, uncertainty, and per-leg, per-repeat MBAR estimates with their errors. |
openfe_convergence.json | Forward and reverse estimates, overlap matrices and eigenvalues, replica transition matrices, replica state histories, and equilibration, production and selected iteration indices. |
openfe_resolved_settings.json | The complete OpenFE protocol settings used for the run. |
openfe_protocol.json, openfe_protocol_dag.json, openfe_protocol_dag_result.json, openfe_protocol_result.json | OpenFE's own serializations of the protocol, its task graph and its results, which OpenFE can reload. |
openfe_provenance.json | OpenFE release, package versions, environment lock and a digest of the submitted ligand. |
vacuum.nc, solvent.nc and checkpoints | Multistate trajectories and energies for each leg and repeat. |
mbar_overlap_matrix.png, replica_exchange_matrix.png, forward_reverse_convergence.png, replica_state_timeseries.png | OpenFE's analysis plots for each leg and repeat. |
structural_analysis.npz, ligand_RMSD.png | Symmetry-corrected ligand RMSD for each solvent leg state over the trajectory. OpenFE does not run structural analysis for the vacuum leg. |
| Equilibration files | Minimized and equilibrated structures (.pdb), the equilibration trajectory (.xtc) and its log. |
Files from different repeats and legs keep OpenFE's unit folder names in their paths. If a run stops at its runtime limit, openfe_runtime_budget_summary.json records why, and the files written before the stop are returned without an estimate.
A negative ΔG (kcal/mol) means the molecule is more stable in water than in vacuum. Polar molecules such as alcohols and amides are usually negative by several kcal/mol, while alkanes and many halogenated aromatics are near zero or positive. For ethanol, the experimental value in the FreeSolv database is −5.0 kcal/mol.
The value is a force field prediction. Its accuracy depends on the molecule, the OpenFF force field, the AM1-BCC charges, the water model and sampling. Compare molecules calculated with the same settings rather than reading a single value as an experimental measurement.
± (kcal/mol) is the spread between independent repeats. It measures how reproducible the sampling is, not how close the value is to experiment. With one repeat there is nothing to compare, and OpenFE reports 0; use three repeats whenever the uncertainty matters. A small spread does not prove convergence, because three repeats can agree while all being undersampled, so read it together with the convergence checks below. Per-repeat MBAR errors for each leg are in openfe_ahfe_results.json.
OpenFE's analysis guidance, based on Klimovich, Shirts and Mobley, gives three checks for each leg and repeat:
mbar_overlap_matrix.png, elements next to the diagonal should be at least 0.03, so that each state overlaps its neighbors and the end states are connected.forward_reverse_convergence.png, forward and reverse estimates should agree within error before all the data is used.replica_exchange_matrix.png and replica_state_timeseries.png, replicas should move through all states without getting stuck in groups.Minimum MBAR overlap in the table is a different, single-number summary: the overlap scalar OpenFE reports for each leg (one minus the largest nontrivial eigenvalue of the overlap matrix), minimized across legs and repeats. Values near 0 mean some states are poorly connected and the estimate should not be trusted; check the matrix plot of the leg with the lowest value. Forward/reverse complete is false when OpenFE could not run that analysis for a repeat, or could not estimate some of its fractions (those fractions are null in openfe_convergence.json), which usually happens with very short runs.
Test and Short runs often fail these checks. If they do, rerun with Production before using the number.
A Production run of phenol with 3 repeats and TIP3P returns one row. Read it in this order: check Forward/reverse complete is true and Minimum MBAR overlap is well above 0; open the overlap and forward/reverse plots for each leg; then read ΔG (kcal/mol) with its ±. If one repeat differs from the others in openfe_ahfe_results.json, its plots usually show which leg did not converge.
Molecular dynamics, minimization, and energy calculations with classical or ML potentials
Calculate binding free energies with MM/PBSA and MM/GBSA
Flexible-ligand molecular docking with grid-based scoring
Enhanced molecular docking with custom scoring functions and fast minimization
Predict biomolecular complex structures and binding affinities for proteins, ligands, DNA, and RNA.
Structure and binding prediction for proteins, DNA, RNA and ligands.
Low-memory Boltz inference for biomolecular structure and affinity prediction
Predict quantum-informed molecular properties from 3D structures.
Run protein molecular dynamics or your own prepared GROMACS system.
Turn an IC50 into pIC50 and an assay-independent Ki, with the equation and every substituted value shown.