TY - BOOK
T1 - Towards Automated and Accurate Predictions of Chemical Reactivity Using Semi-Empirical Quantum Chemistry
AU - Kromann, Jimmy Charnley
PY - 2017
Y1 - 2017
N2 - The ability to use computational tools and methods to predict how molecules react and interact is valuable as chemical space is big and exploring it experimentally is both expensive and time-consuming. Semi-empirical quantum chemistry offers a potential alternative to conventional computational methods by being both fast and accurate. In this thesis I present semi-empirical approaches to solving bio-chemical problems, both with success and also with a great deal of shortcomings. In this work, semi-empirical methods were used in combination with solvation models to predict pKa of common bio-chemical groups and was found in relative good agreement with conventional methods, being within 1 pH unit mean-unsigned-error. As a continuation of the pKa studies, the tool was developed as a method for predicting the regioselectivity of electrophilic aromatic substitution reactions using the combination of semi-empirical quantum chemistry and a continuum solvation model. By prototonating aromatic C-H carbon atoms and identifying the one with the lowest free energy as the most nucleophilic center, it was found that this correlated with the regio-chemical outcome of 96% of more than 500 literature examples. A collection of barrier heights and reaction enthalpies were studied for a total of five enzymes showing semi-empirical methods differ significantly from conventional quantum chemistry methods. The different behavior between SQM and ab initio was further illustrated by computing reaction enthalpies for 25 reactions using the connectivity-based hierarchy (CBH) protocol for accurate results. Using the systematic error cancellation with ab initio methods greatly decreases the error. But for NDDO-based semi-empirical methods the increase in accuracy is very modest due to the random errors in predicted of heats of formation. The semi-empirical methods of GAMESS was extended to include d-type integrals in the MNDO approximation and interfaced with continuum solvation model. This interface was benchmarked with SMD and found that the parameters of this solvation model did not transfer well to semi-empirical methods. With the new implementation the SMD parameters were optimized to give an accuracy similar to conventional methods. Because of the computational speed that semi-empirical methods offer, it is possible to run orders of magnitude more chemical compounds in high-throughput setups than conventional methods. This requires the setup to be automatic which is possible using cheminformatics tools.
AB - The ability to use computational tools and methods to predict how molecules react and interact is valuable as chemical space is big and exploring it experimentally is both expensive and time-consuming. Semi-empirical quantum chemistry offers a potential alternative to conventional computational methods by being both fast and accurate. In this thesis I present semi-empirical approaches to solving bio-chemical problems, both with success and also with a great deal of shortcomings. In this work, semi-empirical methods were used in combination with solvation models to predict pKa of common bio-chemical groups and was found in relative good agreement with conventional methods, being within 1 pH unit mean-unsigned-error. As a continuation of the pKa studies, the tool was developed as a method for predicting the regioselectivity of electrophilic aromatic substitution reactions using the combination of semi-empirical quantum chemistry and a continuum solvation model. By prototonating aromatic C-H carbon atoms and identifying the one with the lowest free energy as the most nucleophilic center, it was found that this correlated with the regio-chemical outcome of 96% of more than 500 literature examples. A collection of barrier heights and reaction enthalpies were studied for a total of five enzymes showing semi-empirical methods differ significantly from conventional quantum chemistry methods. The different behavior between SQM and ab initio was further illustrated by computing reaction enthalpies for 25 reactions using the connectivity-based hierarchy (CBH) protocol for accurate results. Using the systematic error cancellation with ab initio methods greatly decreases the error. But for NDDO-based semi-empirical methods the increase in accuracy is very modest due to the random errors in predicted of heats of formation. The semi-empirical methods of GAMESS was extended to include d-type integrals in the MNDO approximation and interfaced with continuum solvation model. This interface was benchmarked with SMD and found that the parameters of this solvation model did not transfer well to semi-empirical methods. With the new implementation the SMD parameters were optimized to give an accuracy similar to conventional methods. Because of the computational speed that semi-empirical methods offer, it is possible to run orders of magnitude more chemical compounds in high-throughput setups than conventional methods. This requires the setup to be automatic which is possible using cheminformatics tools.
UR - https://rex.kb.dk/permalink/f/h35n6k/KGL01012063079
M3 - Ph.D. thesis
BT - Towards Automated and Accurate Predictions of Chemical Reactivity Using Semi-Empirical Quantum Chemistry
PB - Department of Chemistry, Faculty of Science, University of Copenhagen
ER -