Optimization of Semiempirical Quantum Mechanical Methods for in Silico Drug Design Doctoral thesis Kristian Kříž The subject of this thesis is the optimization of semiempirical quantum mechanical methods (SQM) for their use in in silico drug design. The thesis covers two topics - COSMO2 solvation model optimization part and PLF547, PLA15 dataset development part. The first part is devoted to the optimization of COSMO solvation model by addition of a nonpolar term and reparametrization of the model for SQM methods PM6 and PM7. We have shown that the accuracy of the resulting "COSMO2" optimized model improved on all the tested datasets and we have compared it to other selected SQM solvation models. The method has also been tested on the protein ligand complexes as a part of a scoring function, where it provides better preditction of binding affinity of drug candidates for their target protein. The second part of the thesis describes the construction of datasets for noncovalent interactions aimed speicificly to represent an environment of an enyzme active site complexed with a ligand with reliable benchmark values of interaction energies in vacuum and solvent (water). The developed PLF547 and PLA15 datasets are suitable for testing and development of methods for the use in drug design. We have...

Advances in computing capabilities have facilitated the application of quantum mechanical methods to increasingly larger and more complex chemical systems, including weakly interacting and biologically relevant species. One such ab initio-based composite methodology, the correlation consistent composite approach (ccCA), has been shown to be reliable for the prediction of enthalpies of formation and reaction energies of main group species in the gas phase to within 1 kcal mol-1, on average, of well-established experiment, without dependence on experimental parameterization or empirical corrections. In this collection of work, ccCA has been utilized to determine the proton affinities of deoxyribonucleosides within an ONIOM framework (ONIOM-ccCA) and to predict accurate enthalpies of formation for organophosphorus compounds. Despite the complexity of these systems, ccCA is shown to result in enthalpies of formation to within ~2 kcal mol-1 of experiment and predict reliable reaction energies for systems with little to no experimental data. New applications for the ccCA method have also been introduced, expanding the utility of ccCA to solvated systems and complexes with significant noncovalent interactions. By incorporating the SMD solvation model into the ccCA formulation, the Solv-ccCA method is able to predict the pKa values of nitrogen systems to within 0.7 pKa unit (less than 1.0 kcal mol-1), overall. A hydrogen bonding constant has also been developed for use with weakly interacting dimers and small cluster compounds, resulting in ccCA interaction energies for water clusters and dimers of the S66 set to within 1.0 kcal mol-1 of well-established theoretical values.