Development of tools for evolutionary de novo design of molecules with improved properties
MetadataShow full item record
- Institutt for kjemi 
Design and discovery of molecules using computers and particularly de novo based design has been in practice for many years. In such schemes, the algorithm to design the molecules and precise estimates their properties are relevant for the discovery of structures with desired features. This thesis presents methods and software tools for the design of synthetically realistic structures, di erent approaches to perform necessary computations in a more e cient manner and ways to improve the accuracies of the property estimates. The de novo design method was implemented to discover azobenzene dyes using evolutionary algorithms. The property chosen to be optimised was, the energy of excitation, which is an important property in the eld of molecular electronics and dye-sensitised solar cell (DSSCs) applications. The excitation energies of the molecules from the design program were computed using time-dependent density functional theory (TD-DFT) calculations. Novel derivatives of the dye were discovered with signi cantly lower excitation energies. The property estimates by density functional theory (DFT) and TD-DFT methods greatly vary with the exchange-correlation functional used. With many density functionals and basis-sets available in the literature, it is not practical to perform DFT and TD-DFT computations using all density functionals to identify the most appropriate functional. To address this, an algorithm is presented that helps determine the best functional suitable to compute a certain property of a molecular system. There are many methods available in the literature to compute the properties of molecules with increasing levels of accuracy and computational cost. Starting from less accurate methods based on molecular mechanics to more accurate methods in quantum chemical such as DFT and ab-initio methods. Generally, with increasing accuracy of the method, the computational costs required also increases rapidly. As a way to address this challenge, a software is presented that is designed speci cally to manage high-throughput quantum chemical calculations on massively parallel computing systems. Evolutionary de novo design was applied to discover Coumarin derivatives with improved properties for use in Gr atzel solar cells. Quantitative structure and property relationship models were used to predict relevant properties of the dye. The top structures identi ed from the design scheme were further analysed using DFT computations to support the evidence of their potential use in DSSCs.