Supramolecular

LNCI16

Summary

Performance in predicting host-guest interaction energies for 16 large non-covalent complexes. These include proteins, DNA, and supramolecular assemblies ranging from 380 up to 1988 atoms with diverse interaction motives. 14 complexes are neutral, and two are charged with charges +1 (TYK2) and -2 (FXa).

Metrics

  1. Interaction energy error

For each complex, the interaction energy is calculated by taking the difference in energy between the host-guest complex and the sum of the individual host and guest energies. This is compared to the reference interaction energy, calculated in the same way.

Computational cost

Low: tests are likely to take minutes to run on CPU.

Data availability

Input structures:

  • J. Gorges, B. Bädorf, A. Hansen, and S. Grimme, ‘LNCI16 - Efficient Computation of the Interaction Energies of Very Large Non-covalently Bound Complexes’, Synlett, vol. 34, no. 10, pp. 1135–1146, Jun. 2023, doi: 10.1055/s-0042-1753141.

  • Associated GitHub repository: https://github.com/grimme-lab/benchmark-LNCI16

Note

As described in the above repository, the originally released dataset contained some incorrect values. We have used the corrected values for this benchmark.

Reference data:

  • Same as input data

  • \({\omega}\text{B97X-3c}\) level of theory: a composite range-separated hybrid DFT method with a refitted D4 dispersion correction.

S30L

Summary

Performance in predicting supramolecular binding energies for the S30L set of 30 large host–guest complexes. Systems contain up to ~200 atoms and feature a wide range of interaction motifs, including hydrogen and halogen bonding, π–π stacking, CH–π contacts, nonpolar dispersion, and cation–dipolar interactions. Net charges span −1 to +4, where 8 of the 30 complexes are charged. These ΔE_emp values serve as the benchmark dataset for assessing quantum-chemical methods on large noncovalent complexes.

Metrics

  1. Total binding energy error

For each complex, the binding energy is calculated by taking the difference in energy between the host-guest complex and the sum of the individual host and guest energies. This is compared to the reference binding energy, calculated in the same way.

  1. Charged binding energy error

For each charged complex, the binding energy is calculated by taking the difference in energy between the host-guest complex and the sum of the individual host and guest energies. This is compared to the reference binding energy, calculated in the same way.

  1. Neutral binding energy error

For each neutral complex, the binding energy is calculated by taking the difference in energy between the host-guest complex and the sum of the individual host and guest energies. This is compared to the reference binding energy, calculated in the same way.

Computational cost

Low: tests are likely to take minutes to run on CPU.

Data availability

  • R. Sure and S. Grimme, ‘S30L - Comprehensive Benchmark of Association (Free) Energies of Realistic Host–Guest Complexes’, J. Chem. Theory Comput., vol. 11, no. 8, pp. 3785–3801, Aug. 2015, doi: 10.1021/acs.jctc.5b00296.

  • Stuctures download found in SI

    • TPSS-D3/def2-TZVP geometries of all complexes as Cartesian coordinates.

Reference data:

  • The Supporting Information also provides the benchmark binding energies.

    • These are empirical gas-phase reference values (ΔE_emp), obtained by back-correcting the experimental association free energies with theoretical estimates of vibrational and solvation contributions.

PLF547

Summary

Performance in predicting intramolecular hydrogen bonding and side–chain/backbone contacts in polypeptide fragments for the PLF547 set of 547 shorter peptide-like fragments.

Metrics

  1. MAE

For each complex, the interaction energy is calculated by taking the difference in energy between the ligand-fragment complex and the sum of the individual ligand and protein fragment energies. This is compared to the reference interaction energy.

Computational cost

Low: tests are likely to take minutes to run on CPU.

Data availability

Input structures:

  • K. Kříž and J. Řezáč, “Benchmarking of Semiempirical Quantum-Mechanical Methods on Systems Relevant to Computer-Aided Drug Design”, Journal of Chemical Information and Modeling, vol. 60, no. 3, pp. 1453-1460, 2020, doi: 10.1021/acs.jcim.9b01171

    • DLPNO-CCSD(T)/CBS

Reference data:

Same as input structures

PLA15

Summary

Performance in predicting protein–ligand active-site interaction energies for the PLA15 set of 15 complexes. Systems range from 259 to 584 atoms and contain complete active sites. Ligands contain 37–95 atoms with net charges of −1, 0, or +1 and all contain aromatic heterocycles. Five ligands contain either divalent of tetrahedral sulfur atoms, and four and three of them contain F and Cl atoms, respectively.

Metrics

  1. Total MAE

For each complex, the interaction energy is calculated by taking the difference in energy between the protein-ligand complex and the sum of the individual protein and ligand energies. The MAE is computed by comparing predicted interaction energies to reference interaction energies across all 15 systems.

  1. Pearson’s r²

The squared Pearson correlation coefficient between predicted and reference interaction energies, measuring the proportion of variance in the reference values explained by the model predictions.

  1. Ion-Ion MAE

For each complex where both protein and ligand fragments have non-zero charges, the interaction energy error is calculated. This metric reports the MAE for these ion-ion interaction systems.

  1. Ion-Neutral MAE

For each complex where one fragment (protein or ligand) has a non-zero charge and the other is neutral, the interaction energy error is calculated. This metric reports the MAE for these ion-neutral interaction systems.

Computational cost

Low: tests are likely to take minutes to run on CPU.

Data availability

Input structures:

  • K. Kříž and J. Řezáč, ‘protein ligand - Benchmarking of Semiempirical Quantum-Mechanical Methods on Systems Relevant to Computer-Aided Drug Design’, J. Chem. Inf. Model., vol. 60, no. 3, pp. 1453–1460, Mar. 2020, doi: 10.1021/acs.jcim.9b01171.

  • Structures download found in SI

Reference data:

  • The Supporting Information also provides the interaction energies.

    • The benchmark interaction energies are based on a combination of explicitly correlated MP2-F12 calculations and a DLPNO-CCSD(T) correction