“I have an initial set of hits from a virtual screen, but I want a refined set with fewer false positives and false negatives, validated by more robust, physics-based simulations.”

- Some compounds score well for the wrong reasons (false positives)
- Some true binders are missed due to imperfect sampling or rigid assumptions (false negatives)
- Binding stability and solvent effects are not fully captured
This workflow refines your hit list using a primary refinement chain: Docking → Protein–Ligand MD → ABFE/RBFE. Each step increases realism and confidence for your compounds of interest:
- Docking provides a fast structural sanity check and rescoring method that can be applied to larger library sizes
- Protein-Ligand MD tests whether poses are stable over time in solvent, and with dynamic time conditions as well
- ABFE/RBFE quantifies binding energetics with much higher fidelity than docking
Required
- Hit molecules as SMILES (CSV preferred)
- A protein structure (experimental or predicted) with a defined binding site
- A clear binding pocket definition (from known ligand, site annotation, or docking grid)
- Known reference ligand(s) or co-crystal pose (for benchmarking)
- Experimental activity data (if any)
- Notes on liabilities to avoid (reactive groups, solubility risk, etc.)
- Re-score and Sanity-Check Hits
- If your hit set is large: run Static Docking first for throughput, then downselect based on set binding criteria for your project
- If your hit set is smaller: go straight to Flexible Docking for better pose realism, or to Ensemble Docking for more advanced workflows and protein systems
- Clear, plausible binding poses (no severe clashes)
- Consistent strong scores across top poses (low variation between best and average)
- A pose that makes chemical sense (key interactions, reasonable geometry)
- Validate Binding Pose Stability
- RMSD vs time: a quick rise then plateau (no long-term drift)
- Radius of gyration / SASA: stable protein behavior (no unfolding or destabilization)
- RMSF near pocket residues: no extreme instability; stable binding-site behavior
- Visual check: ligand stays in the pocket and does not flip, drift, or dissociate from the pocket
- Quantify Binding Energetics
- ABFE: “Is binding thermodynamically favorable, and how strong is it?”
- RBFE: “Between similar ligands, which binds better and by how much?”
- Solvent reorganization
- Protein–ligand dynamics
- Entropy and restraint corrections
- More realistic thermodynamics than docking scores
- More favorable (more negative) binding free energies for stronger candidates
- Reasonable, stable energetic components (no obvious artifacts)
- Consistency across methods (TI vs MBAR) when available
- Clear ranking that separates top candidates from the rest
- Final Refinement and Selection
- structurally plausible (Docking)
- dynamically stable (Protein–Ligand MD)
- thermodynamically supported (ABFE/RBFE)
- Select a small number of compounds for synthesis / purchase
- Prepare follow-on hit-to-lead expansion campaigns
- A refined set of hits with significantly fewer false positives
- Better protection against false negatives (through better pose realism and dynamics)
- Ranked candidates supported by higher-fidelity binding energetics
- Compounds ready for experimental validation or lead optimization workflows
- You can move these compounds more confidently into experimentation, knowing that these hypotheses have been rigorously tested through a multi-modal procedure to assess for different binding dynamics.
- After being tested initially, you can utilize generative chemistry workflows to optimize your lead sets using similar methods
Revilico makes hit refinement a single connected workflow: you can start with virtual screening outputs, validate stability with MD, then quantify binding with FEP-grade calculations, all without rebuilding inputs across tools. This produces a refined shortlist that is both computationally grounded and scientifically interpretable, so teams can make better downstream decisions with less wasted synthesis and assay effort.

