Predicting_functional_metabolic_control_mechanisms_using_Graph_Convolutional_Neuronal_Networks

The systematic identification of functional metabolic control mechanisms relies mostly on the integration of data but barely considers the connectivity of the metabolic network. However, recent geometric deep learning approaches show promising performances in the prediction of links in network or graph structures. By using a dream-case in silico network of E. coli’s central metabolism including metabolic and regulatory connections, we first generated data suitable for the prediction of regulatory links between metabolites and enzymes. By merging data with a Graph Convolutional Neuronal Network along the structure of the metabolic network we can classify links in the graph as either ‘regulatory’ or ‘non-regulatory’. Here we show, how such a neuronal network can be trained and also be used to fill knowledge gaps to identify functional metabolic control mechanisms.