Validation Against Gold-Standard Quantum Chemistry
How the RFT engine is tested: standard public benchmarks, gold-standard references, zero fitted parameters.
Quantum interaction energies
The engine is compared against reference interaction energies from high-level quantum chemistry. On small-molecule benchmarks the references are symmetry-adapted perturbation theory (SAPT2+) and near full-configuration-interaction. On the protein-ligand set the reference is CCSD(T) at the complete-basis-set limit.
Small-molecule benchmarks
| Benchmark | Reference | Correlation |
|---|---|---|
| S66 dimers | SAPT2+ | 0.92 |
| A24 | near full-CI | 0.84 |
| S66x8 (528 points) | SAPT2+ | 0.90 |
Rank correlation against the reference energies, computed with zero fitted parameters.
Protein-ligand interaction energies (PLA15)
PLA15 is a set of fifteen protein-ligand active-site complexes with CCSD(T)/CBS reference interaction energies. The table reports Pearson correlation and mean absolute error against that reference, for the RFT engine and for a range of trained machine-learning potentials and a semi-empirical method, all evaluated on the same fifteen structures.
| Method | Correlation | MAE (kcal/mol) | Trained parameters |
|---|---|---|---|
| eSEN | 0.996 | 14.4 | millions |
| GFN2-xTB | 0.992 | 10.6 | semi-empirical |
| UMA-S | 0.991 | 16.3 | millions |
| AIMNet2 | 0.985 | 35.8 | millions |
| RFT engine | 0.92 | 21.7 | none |
| ANI-2x | 0.71 | 73.0 | millions |
| MACE | 0.78 | 112.2 | millions |
PLA15 dataset and CCSD(T)/CBS reference: Kriz and Rezac, Journal of Chemical Information and Modeling, 2020. Comparator predictions: Rowan Scientific pla15-benchmarks (MIT licensed). Correlation and mean absolute error were computed by us on the identical fifteen structures. Every method at or above the RFT engine was trained on millions of quantum-chemistry calculations. The RFT engine uses none.
Docking and screening
For pose ranking and virtual screening, the engine is validated on the DUD-E and DEKOIS 2.0 benchmarks with property-matched decoys, using five-fold cross-validation and Y-scramble null controls. Selected holdout AUC values:
| Target | Family | AUC |
|---|---|---|
| HMGR | Reductase | 0.980 |
| PDE5 | Phosphodiesterase | 0.883 |
| Neuraminidase | Glycosidase | 0.875 |
| MCL1 | Apoptosis regulator | 0.868 |
| ESR1 | Nuclear receptor | 0.858 |
| ADRB2 | GPCR | 0.834 |
| CATL | Cysteine protease | 0.829 |
Holdout-validated on DEKOIS 2.0 with property-matched decoys, five-fold cross-validation, and Y-scramble null controls.
What the validation shows
Across both small molecules and protein-ligand complexes, a single engine with no fitted parameters reproduces gold-standard quantum chemistry and reaches the accuracy tier of methods trained on millions of reference calculations, while using none. Because the engine carries no fitted parameters, every test complex is a genuine prediction rather than a recall of training data.