Four alchemical free-energy engines benchmarked head-to-head on one force field
- Compares Amber thermodynamic integration, GROMACS non-equilibrium TI, OpenMM FEP and BLaDE multisite lambda dynamics using identical Amber force fields via the MolCube-AFES input workflow.
- Tested on a published benchmark set of ligand transformations plus two further applications, against experimental data.
- All four reproduced experimental trends with no statistically significant accuracy differences; choice can rest on throughput, portability and perturbation-network design.
💡 Points Based on the abstract
A tie on accuracy is only a tie on this benchmark: the Wang et al. set is mostly neutral, congeneric, rigid-ish ligands, and everything shared one Amber force field. Before picking an engine, rerun your own hardest series (charged groups, ring changes, flexible linkers) and log GPU-hours per converged edge plus run-to-run spread. That decides it, not the headline parity.