We present an extension to the recent 3OB parametrization of the Density Functional Tight Binding Model DFTB31,2 for biological and organic systems. Parameters for the halogens F, Cl, Br, and I have been developed for use in covalently bound systems and benchmarked on a test set of 106 molecules (the ‘OrgX’ set), using bonding distances, bonding angles, atomization energies, and vibrational frequencies to assess the performance of the parameters. Additional testing has been done with the X40 set of 40 supramolecular systems containing halogens,3 adding a simple correction for the halogen bonds that are strongly overbound in DFTB3. Furthermore, parameters for Ca, K, and Na as counterions in biological systems have been created. To benchmark geometries as well as ligand binding energies a test set ‘BioMe’ of 210 molecules has been created that cover coordination to various functional groups frequently occurring in biological systems. The new DFTB3/3OB parameter set outperforms DFT calculations with a double-ζ basis set in terms of energies and can reproduce DFT geometries, with some minor deviations in bond distances and angles due to the use of a minimal basis set.

We present the parametrization and benchmark of long-range corrected second-order density functional tight binding (DFTB), LC-DFTB2, for organic and biological molecules. The LC-DFTB2 model not only improves fundamental orbital energy gaps but also ameliorates the DFT self-interaction error and overpolarization problem, and further improves charge-transfer excited states significantly. Electronic parameters for the construction of the DFTB2 Hamiltonian as well as repulsive potentials were optimized for molecules containing C, H, N, and O chemical elements. We use a semiautomatic parametrization scheme based on a genetic algorithm. With the new parameters, LC-DFTB2 describes geometries and vibrational frequencies of organic molecules similarly well as third-order DFTB3/3OB, the de facto standard parametrization based on a GGA functional. LC-DFTB2 performs well also for atomization and reaction energies, however, slightly less satisfactorily than DFTB3/3OB.

Despite decades of investigations, the principal mechanisms responsible for the high affinity and specificity of proteins for key physiological cations K+, Na+, and Ca2+ remain a hotly debated topic. At the core of the debate is an apparent need (or lack thereof) for an accurate description of the electrostatic response of the charge distribution in a protein to the binding of an ion. These effects range from partial electronic polarization of the directly ligating atoms to long-range effects related to partial charge transfer and electronic delocalization effects. While accurate modeling of cation recognition by metalloproteins warrants the use of quantum-mechanics (QM) calculations, the most popular approximations used in major biomolecular simulation packages rely on the implicit modeling of electronic polarization effects. That is, high-level QM computations for ion binding to proteins are desirable, but they are often unfeasible, because of the large size of the reactive-site models and the need to sample conformational space exhaustively at finite temperature. Several solutions to this challenge have been proposed in the field, ranging from the recently developed Drude polarizable force-field for simulations of metalloproteins to approximate tight-binding density functional theory (DFTB). To delineate the usefulness of different approximations, we examined the accuracy of three recent and commonly used theoretical models and numerical algorithms, namely, CHARMM C36, the latest developed Drude polarizable force fields, and DFTB3 with the latest 3OB parameters. We performed MD simulations for 30 cation-selective proteins with high-resolution X-ray structures to create ensembles of structures for analysis with different levels of theory, e.g., additive and polarizable force fields, DFTB3, and DFT. The results from DFT computations were used to benchmark CHARMM C36, Drude, and DFTB3 performance. The explicit modeling of quantum effects unveils the key electrostatic properties of the protein sites and the importance of specific ion-protein interactions. One of the most interesting findings is that secondary coordination shells of proteins are noticeably perturbed in a cation-dependent manner, showing significant delocalization and long-range effects of charge transfer and polarization upon binding Ca2+.

We combine the approximate density-functional tight-binding (DFTB) method with unsupervised machine learning. This allows us to improve transferability and accuracy, make use of large quantum chemical data sets for the parametrization, and efficiently automatize the parametrization process of DFTB. For this purpose, generalized pair-potentials are introduced, where the chemical environment is included during the learning process, leading to more specific effective two-body potentials. We train on energies and forces of equilibrium and nonequilibrium structures of 2100 molecules, and test on ∼130 000 organic molecules containing O, N, C, H, and F atoms. Atomization energies of the reference method can be reproduced within an error of ∼2.6 kcal/mol, indicating drastic improvement over standard DFTB.