Free Energy Calculations Usage Guide

Overview

Perform alchemical free-energy calculations using OpenFE or a commercial workflow. Accuracy is system- and protocol-dependent, so validate mappings, sampling, network consistency, and prospective performance rather than promising a universal RMSE.

Prerequisites

mamba create -n openfe -c conda-forge openfe=1.7 alchemlyb pymbar
conda activate openfe

Install OpenFE from conda-forge as documented by the project. The unrelated package named openfe on PyPI is not the Open Free Energy toolkit.

Quick Start

Tell the AI agent what to do: - "Run OpenFE RBFE between lig1 and lig2 in this receptor" - "Compute ABFE for a single ligand bound to my receptor" - "Run MM/GBSA endpoint scoring on docked poses" - "Check FEP cycle closure for my 5-ligand perturbation graph" - "Boltz-2 affinity first-pass on 100 candidates"

Example Prompts

Lead optimization RBFE

"Run OpenFE RBFE on a congeneric series for kinase X. Start from versioned protocol defaults, inspect mappings, and extend sampling from overlap, replicate, and closure-residual diagnostics."

Quick MM/GBSA screen

"Run MM/GBSA on the top 100 Vina-docked poses for receptor.pdb. Rank by delta-G_binding. Output to results.csv with sd."

Boltz-2 affinity triage

"Boltz-2 affinity prediction on 1000 SMILES against receptor.pdb. Rank by predicted affinity. Output top 20 for FEP follow-up."

ABFE for single ligand

"Compute OpenFE ABFE for ligand.sdf bound to receptor.pdb using the documented restraint workflow. Record all settings and extend each leg from convergence diagnostics."

What the Agent Will Do

1. Set up perturbation: ligand1 -> ligand2 (RBFE) or ligand -> uncoupled (ABFE).
2. Generate and inspect atom mappings (Kartograf is the OpenFE 1.7 CLI default; LOMAP is supported).
3. Start from the release-matched lambda schedule and modify it only from overlap and protocol diagnostics.
4. Equilibrate and run production sampling, extending from replicate and time-series convergence evidence.
5. Analyze with MBAR/BAR via alchemlyb.
6. Compute cycle closure error; report delta-delta-G ± SD.
7. Validate convergence using replicate agreement, overlap, exchange, and signed closure residuals with propagated uncertainty.

Tips

• OpenFE 1.7's versioned documentation shows OpenFF 2.1.1 for ligands; inspect and record the actual installed protocol settings.
• OpenFE's default RBFE Hamiltonian replica exchange is not REST2. Use enhanced sampling only when the selected engine and protocol document it.
• Interpret signed cycle-closure residuals with edge uncertainties, shared-edge correlations, and replicate diagnostics; RMS requires multiple specified cycles.
• Estimate ABFE/RBFE cost from the explicit protocol rather than a fixed multiplier.
• Use MM/GBSA only where a matched benchmark supports the intended ranking decision.
• The Boltz-2 preprint reports Pearson 0.66 on its FEP benchmark subset and at least 1000-fold lower computational cost; treat this as benchmark-specific and confirm top candidates with FEP.

Related Skills

• chemoinformatics/virtual-screening - Source docking poses for FEP input
• chemoinformatics/pose-validation - PoseBusters before FEP setup
• chemoinformatics/conformer-generation - Ligand 3D for FEP
• chemoinformatics/molecular-standardization - Standardize before FEP
• chemoinformatics/ml-docking-rescoring - Boltz-2 affinity alternative
• chemoinformatics/qsar-modeling - Surrogate models for screening