Literature mining supports a next-generation modeling approach to predict cellular byproduct secretion

•Literature mining is used to create a database of E. coli cell factory strains.•Common design strategies are identified.•Designs are simulated in six historical genome-scale models, including the ME-model.•Model predictions have improved as models have expanded in size and scope.•ME-model predictions can be improved by parameterizing kinetics.

The metabolic byproducts secreted by growing cells can be easily measured and provide a window into the state of a cell; they have been essential to the development of microbiology, cancer biology, and biotechnology. Progress in computational modeling of cells has made it possible to predict metabolic byproduct secretion with bottom-up reconstructions of metabolic networks. However, owing to a lack of data, it has not been possible to validate these predictions across a wide range of strains and conditions. Through literature mining, we were able to generate a database of Escherichia coli strains and their experimentally measured byproduct secretions. We simulated these strains in six historical genome-scale models of E. coli, and we report that the predictive power of the models has increased as they have expanded in size and scope. The latest genome-scale model of metabolism correctly predicts byproduct secretion for 35/89 (39%) of designs. The next-generation genome-scale model of metabolism and gene expression (ME-model) correctly predicts byproduct secretion for 40/89 (45%) of designs, and we show that ME-model predictions could be further improved through kinetic parameterization. We analyze the failure modes of these simulations and discuss opportunities to improve prediction of byproduct secretion.

Graphical abstractFigure optionsDownload full-size imageDownload high-quality image (240 K)Download as PowerPoint slide