Automated generation of reaction network hypotheses for complex feedstocks

The advent of greater environmental concern has limited the industrial use of conventional fossil fuel feedstocks, leading to a shift towards bio-based feedstocks that are equally complex. These alternative sources are typically characterized by a multitude of components and variability in composition. This poses a challenge in developing reliable models of the reactions in the system, which are necessary to maintain optimal downstream requirements and requires expert domain knowledge. In this work, we aim to establish an automated protocol to infer reaction network hypotheses for such complex systems based on spectroscopic information, tensor decomposition and machine learning frameworks thereby obviating the need for human expert knowledge. The routine is validated on the processing of biomass and its model compounds.