Prediction of Biodegradation Pathways

The majority of organic matter released into the environment is degraded by microorganisms. This does not happen in a single step, but in a sequence of reactions, which form a biodegradation pathway. If a parent compound is degraded, it is transformed to several biodegradation products, which may be more or less dangerous for environmental and human health than the parent compound.

The University of Minnesota Biocatalysis/Biodegradation Database (UM-BBD) is presently the largest database with experimentally determined biodegradation pathways. Although it contains already more than 600 compounds and almost 100 pathways, it would be very valuable to to have some information about the environmental fate of the other approximately 10 million organic compounds currently known.

As the UM-BBD is too complex for an in depth analysis by human experts, we will develop and apply machine learning programs to extract knowledge about biodegradation mechanisms from this data. This is a completely new problem for machine learning. We will start therefore to develop a representation of chemical structures and reactions, which is adequate for the information in the UM-BBD, usable by machine learning programs and interpretable by biodegradation experts. Based on this representation we will adopt and develop machine learning techniques for this special learning problem.

The application of these techniques to the UM-BBD shall result in a model for biodegradation pathways. This model can be used to predict the biodegradation products from a parent compound, but it will also help to elucidate the biochemical mechanisms for biodegradation. Therefore it can assist in the development of new, environmentally safe technologies (e.g. biocatalysis) and the optimization of environmental technologies (e.g. activated sludge treatment, biofilters, bioremediation, ...).