Comparing Empirical and Physics-Based Models of Intermolecular Dispersion and Repulsion Energies.

Understanding molecular recognition, crystal packing and biological processes including protein-ligand binding, protein folding, and macromolecular structure all depend on accurate modeling of noncovalent interactions. In this study, we introduce and evaluate two complementary models─empirical and physically grounded ─for computing exchange-repulsion, and dispersion energies. Once combined, these two terms form van der Waals (vdW) potentials allowing the computation of total vdW interaction energies. A set of 21 atom-type-specific vdW parameters of a buffered 7-14 potential was derived via a least-squares optimization using the SciPy library and implemented in the MoProSuite software. The empirical model yields highly accurate interaction energies, exhibiting a strong correlation = 0.990 with reference symmetry-adapted perturbation theory (SAPT) values from the extensive NENCI-2021 data set comprising 6000 diverse small-molecule dimers. The physical model has been implemented in MoProViewer, the molecule and molecular properties viewer of the MoProSuite software package. It utilizes transferred electron density and anisotropic atomic polarizabilities from the ELMAM2 electron density database to estimate exchange-repulsion and dispersion energies, respectively. The vdW energies determined using the physical model exhibit a significant correlation, = 0.956, with SAPT data. A distance-dependent analysis showed that correlation improves with increasing dimer separation (0.80 to 1.1 factor from equilibrium distance). Together, these results demonstrate that the empirical model, supported by optimized parameters, offers a computationally efficient and accurate alternative to high-level quantum mechanical methods. The physical model further emphasizes the significance of accurate electron density and polarizability parameters for correct vdW interaction estimation, without the need for ad-hoc fitted atomic parameters. Validation against a separate benchmark of protein side chain-side chain interactions (SSI) confirmed that both models effectively capture key noncovalent interactions in biologically important molecular systems.