Solvation Thermodynamics BENCHMARK
Scope: Aqueous solubility validated on 1,128 ESOL compounds. Four-term solvation model from first principles. Full methodology published.
Scope: Aqueous solubility validated on 1,128 ESOL compounds. Four-term solvation model from first principles. Full methodology published.
How FluxMateria predicts solvation properties
Solvation free energies are computed from four physically-motivated terms, each derived from molecular geometry. No training data required.
Energy to create a solute-sized cavity in the solvent. Scales with molecular surface area.
van der Waals interactions between solute and surrounding solvent molecules.
Charge-dipole and dipole-dipole interactions from molecular polarity.
Directional H-bond donor/acceptor contributions from FLUX geometry.
1,128 compounds from the Delaney aqueous solubility dataset. Standard benchmark for solubility prediction models. All experimental logS values.
Additional validation on AqSolDB (0.84 logS MAE), confirming generalization beyond the primary ESOL test set.
Performance on standard benchmark datasets
| Dataset | Compounds | logS MAE | RMSE | Bias | Status |
|---|---|---|---|---|---|
| ESOL (Delaney) | 1,128 | 0.71 | 0.913 | +0.024 | PASS |
| AqSolDB | 9,982 | 0.84 | 1.08 | -0.11 | PASS |
ESOL target: MAE < 0.8 logS units. Both datasets pass the commercial-grade accuracy threshold.
How FluxMateria compares to solvation models and ML approaches
| Method | logS MAE | Speed | Training Data | Parameters |
|---|---|---|---|---|
| FluxMateria | 0.71 | ~10 ms | None | 0 fitted |
| ESOL Consensus Model | 0.75 | ~1 ms | Fitted descriptors | 4 fitted |
| Random Forest (fingerprints) | 0.69 | ~5 ms | Training set required | Thousands |
| SMD (Implicit Solvation) | 0.6-1.0 kcal/mol* | Minutes-hours | None | Atomic radii fitted |
| COSMO-RS | 0.5-0.8 kcal/mol* | Minutes | None | Several fitted |
* SMD and COSMO-RS report hydration free energy (kcal/mol), not logS directly. Values are approximate conversions for comparison purposes. FluxMateria achieves competitive accuracy with instant predictions.
30+ solvents supported across multiple classes
Water, methanol, ethanol, isopropanol, n-butanol, acetic acid, formic acid
DMSO, DMF, acetonitrile, acetone, NMP, THF
Hexane, toluene, benzene, cyclohexane, diethyl ether, DCM, chloroform
Carbon tetrachloride, nitromethane, pyridine, 1,4-dioxane, ethyl acetate, MTBE
Thermodynamic relationships verified across predictions
Where predictions are most and least reliable
Primary data sources for experimental validation
Predict solvation thermodynamics across multiple solvents with full interpretability.