The set contains 23 hydrogen bonds featuring all possible combionations of the most common donor and acceptor groups; 23 dispersion-dominated complexes covering pi-pi, aliphatic-aliphatic and pi-aliphatic interactions; and 20 complexes with mixed electrostatic/dispersion interaction. The benchmark CCSD(T)/CBS interaction energies are based on MP2/CBS calculations in aug-cc-pVTZ and aug-cc-pVQZ basis sets and CCSD(T) correction calculated in aug-cc-pVDZ basis set.
Full text of the paper is available: S66.pdf
More data for download

Geometries were constructed by scaling the closest intermolecular distance in the complexes by a factor of 0.9, 0.95, 1.0, 1.05, 1.1, 1.25, 1.5 and 2.0, starting from MP2/cc-pVTZ geometry. The benchmark CCSD(T)/CBS interaction energies are based on MP2/CBS calculations in aug-cc-pVTZ and aug-cc-pVQZ basis sets and CCSD(T) correction calculated in aug-cc-pVDZ basis set.
Full text of the paper is available: S66.pdf
More data for download

The set was developed for parameterization of hydrogen bonding correction for semiempirical QM methods. Geometries were constructed by scaling the closest intermolecular distance in the complexes by a factor of 0.9, 0.95, 1.0, 1.05, 1.1, 1.25, 1.5 and 2.0, starting from MP2/cc-pVTZ geometry. The benchmark CCSD(T)/CBS interaction energies are based on MP2/CBS calculations in aug-cc-pVTZ and aug-cc-pVQZ basis sets and CCSD(T) correction calculated in aug-cc-pVDZ basis set.

Geometries of noncovalent complexes from the S22 dataset were displaced along intermolecular axis, forming one shortened and three elongated (0.9, 1.2, 1.5 and 2.0 times the original intermolecular distance) structures. The dataset also includes the original geometry (labeled 1.0). CCSD(T)/CBS interaction energies consistent with the original S22 work have been calculated.

S22 set consists of small to relatively large (30 atoms) complexes of common molecules containing only C, N, O and H, and single, double and triple bonds. Most typical noncovalent interactions, such as hydrogen bonds (XHY), dispersion interactions (stacked parallel, T-shaped), and mixed electrostatic-dispersion interactions are represented. A total of 22 complexes are divided into three subgroups: (i) hydrogen bonded complexes; (ii) complexes with predominant dispersion stabilization; (iii) mixed complexes in which electrostatic and dispersion contributions are similar in magnitude. Cunterpoise-corrected gradient optimization was used to obtain the geometries. The smallest complexes were optimized by the CCSD(T) method (numerical gradients) using cc-pVTZ and cc-pVQZ basis sets without counterpoise correction. We believe that our S22 set will manage to represent non-covalent interactions in biological molecules in a balanced way and that it will help to design and test fast computational tools for biologically oriented applications.

A detailed quantum chemical study on five peptides (WG, WGG, FGG, GGF and GFA) containing the residues phenylalanyl (F), glycyl (G), tryptophyl (W) and alanyl (A)—where F and W are of aromatic character—is presented. When investigating isolated small peptides, the dispersion interaction is the dominant attractive force in the peptide backbone–aromatic side chain intramolecular interaction. Consequently, an accurate theoretical study of these systems requires the use of a methodology covering properly the London dispersion forces. For this reason we have assessed the performance of the MP2, SCS-MP2, MP3, TPSS-D, PBE-D, M06-2X, BH&H, TPSS, B3LYP, tight-binding DFT-D methods and ff99 empirical force field compared to CCSD(T)/complete basis set (CBS) limit benchmark data. All the DFT techniques with a ‘-D’ symbol have been augmented by empirical dispersion energy while the M06-2X functional was parameterized to cover the London dispersion energy. For the systems here studied we have concluded that the use of the ff99 force field is not recommended mainly due to problems concerning the assignment of reliable atomic charges. Tight-binding DFT-D is efficient as a screening tool providing reliable geometries. Among the DFT functionals, the M06-2X and TPSS-D show the best performance what is explained by the fact that both procedures cover the dispersion energy. The B3LYP and TPSS functionals—not covering this energy—fail systematically. Both, electronic energies and geometries obtained by means of the wave-function theory methods compare satisfactorily with the CCSD(T)/CBS benchmark data.Conformer energies are set relative to average for given peptide in each method.