Research Project

Reaction-Driven Chemical Spaces

Combinatorial chemical fragment spaces define a virtual yet synthetically accessible chemical space by systematically combining molecular fragments through defined reactions. Within the DFG SPP 2363 "Molecular Machine Learning", these spaces are to be expanded using data-driven methods. A central challenge is the quality of publicly available reaction data: relevant datasets are often proprietary or noisy. The current focus therefore lies on developing tools for the standardization, comparison, and curation of public reaction data. In the long term, the fundamental problem of missing negative data in reaction prediction is to be addressed, for example through NLP methods applied to electronic lab notebooks or by combining active learning with automated laboratory systems.