The “chemical space” concept refers to the feature space of a given type of molecule, e.g. the set of all organic compounds up to a certain size, which adhere to simple design rules regarding chemical stability. Researchers in this field have focused on finding structure-property relationships within chemical spaces, e.g. in order to estimate the pharmaceutical activity of an unknown compound. We propose a new perspective of how chemical spaces can be explored, focused on the chemical reactions that connect molecules. This transforms the spaces from simple enumerations to networks of molecules and reactions. This is invaluable for catalysis research, e.g. when considering the elementary steps leading from syngas (H2+CO) to more complex (fuel) molecules like methanol. The goal of the project is to combine the expertise of theoretical chemists in modelling chemical reactions with that of data scientists in analysing complex graphs. It is unfeasible to exhaustively determine every reaction rate in a network, which may consist of billions of molecules. To overcome this issue, a hierarchy of increasingly accurate methods for determining reaction energetics including measures of uncertainty will be developed in the Reuter group. In the Günnemann group, algorithms for adaptive exploration of reaction networks will be developed. Ultimately, we aim to find likely paths through reaction networks, from given educts to desired products. Importantly, we aim to minimize both the computational effort and the expected error, by taking advantage of the hierarchy of methods mentioned above.
Finding the Right Bricks for Molecular Lego: A Data Mining Approach to Organic Semiconductor Design Christian Kunkel, Christoph Schober, Johannes T. Margraf, Karsten Reuter, Harald Oberhofer Chem. Mater. 2019, 31, 3, 969-978 – DOI: 10.1021/acs.chemmater.8b04436