Characterization and engineering of microbial metabolism by using bio big data and machine learning

Abstract

Understanding and optimal engineering of a production host’s metabolism have now been better facilitated by the increasing availability of bio big data and the use of machine learning methods. Here, a suite of machine learning models will be presented that can facilitate several stages of metabolic engineering. Representative examples include DeepEC, DeepRFC and DeepMGR. DeepEC is a deep learning model that predicts enzyme commission (EC) numbers of a protein sequence with high precision in a high-throughput manner. Meanwhile, DeepRFC examines the feasibility of a large number of retrobiosynthesis-derived enzymatic reactions. Retrobiosynthesis helps systematically design novel biosynthetic pathways for the production of a target chemical, but often generates a very large number of candidate reactions, which makes experimental validation very challenging. DeepRFC can facilitate implementation of retrobiosynthesis by substantially reducing a large number of candidate enzymatic reactions. Finally, DeepMGR predicts changes in the expression level of genes in a microbial cell in response to environmental conditions, in particular medium composition. Further ongoing studies will be discussed. Precise definition of problems in metabolic engineering, generation of relevant biological datasets, and development of computational models that can address the problems will innovate our approaches to metabolic engineering.