Artificial Intelligence Discovers Enzymes: Decoding the Missing Link in Metabolism and Making Drugs | Kobe Research


Associate Professor Christopher J. Vavricka, Division of Science, Kobe University, Graduate School of Technology and Innovation, Assistant Professor Shunsuke Takahashi, School of Science and Technology, Tokyo Electric University, Michihiro Araki, Deputy Director, AI Health and Medicine Research Center, Biomedicine, Kobe University, National Institute of Technology Innovation, Health and Nutrition A research team led by Professor Masahisa Hasunuma of the Center for Advanced Bioengineering Research has succeeded in producing a plant-derived pharmaceutical raw material microbial and metabolic by developing a machine learning predictive model capable of discovering and linking unknown enzymes project. In the future, it is expected to accelerate the biological production of various useful substances, functional materials and general chemicals. The results of the study were published March 16 in the British Journal of Science. Nature Communications It was published on.


  • In recent years, with the progress of synthetic biology, the microbial fermentation production of plant-derived pharmaceutical raw materials is highly anticipated.
  • When targeting BIA, which is widely used as a raw material for analgesics, the problem is that some of the enzymes that make up the metabolic pathways are unknown.
  • developed and designed a machine learning predictive model to solve the enzyme discovery problem (Design) – build (Seconduild) – evaluate (Tonest)-learn (largeLink to Earn’s DBTL Workflow).
  • He discovered an unknown enzyme (the missing link) and successfully produced BIA via E. coli.
  • The AI ​​x bio method developed in this study can be applied to the production of various pharmaceutical raw materials, functional materials, and general chemicals, and is expected to contribute to SDGs through environmentally friendly bioproduction.

Research Background

Many of the world’s medicines are made from compounds extracted from plants. Since these compounds (raw materials for medicinal products) are abundant in plants, obtaining them requires large-scale cultivation and industrial processing, which poses an environmental and economic burden. Benzylisoquinoline alkaloids (BIAs) are widely used as raw materials for analgesics, and this situation has been a long-standing problem in large-scale production.

On the other hand, recent advances in biotechnology have been remarkable. By introducing and expressing plant-derived genes in microorganisms, the metabolic pathways of plants can be realized in microorganisms, producing useful substances that microorganisms could not otherwise produce. already become. By cultivating a large number of easy-to-cultivate microorganisms and efficiently fermenting and producing useful substances, the burden and economy of the production process on the environment can be reduced. This approach, called synthetic biology approach, is a new trend in biotechnology-based manufacturing (bioproduction). BIA is also expected to apply this approach, but there are technical challenges.

To produce BIA by microorganisms, a long metabolic pathway consisting of many enzymatic reactions needs to be constructed, but the problem is that some plant-derived enzymes are unknown. This “missing link” is a recurring problem in bioproduction using synthetic biology approaches, and in many cases cannot be solved even if it requires considerable time and effort from molecular biologists. In this study, we developed a machine learning algorithm for predicting enzymatic reactions and successfully discovered the enzymes required for BIA production by applying the predictive model.

research content

In this study, we constructed a DBTL workflow consisting of design, build, test, and learn to search for unknown enzymes (Figure 1). In this new concept, chemical reactions are designed by information science, recombinases are produced by genetic engineering, enzyme functions are assessed by metabolic engineering, and enzymes are searched by machine learning. By turning this cycle, the enzymes that catalyze the unknown reactions are narrowed down and eventually discovered.

I found that this research develops a machine learning algorithm that predicts enzymatic reactions and builds new predictive models by plugging it into a DBTL workflow that produces an enzyme that produces opioid towns Pain medication precursor chemicals (BIAs). The breakthrough of this research lies in the development of real-time machine learning, in addition to the data processing of the design steps, the artificial intelligence prediction model is also established using the newly obtained data from laboratory experiments. The first study to demonstrate its effectiveness.

next deployment

Our goal is to use the obtained enzyme to mass-produce BIA. The AI ​​x bio method developed in this study can be applied to the production (bioproduction) of various pharmaceutical raw materials, functional materials, and general chemicals, and makes a great contribution to SDGs through environmentally friendly processes. .


This research is a research and development project of the New Energy and Industrial Technology Development Institute of the National Research and Development Corporation “Development of High-Performance Product Production Technology Using Plants and Other Organisms/Development of High-Performance Product Production Technology Using Microorganisms” and “Realization of Carbon Cycle”. Supported by “Development of Bio-Derived Product Production Technologies to Accelerate/Create New Enzyme Resources from Database Spaces”.

Thesis information

“Machine Learning Discovers Missing Links Mediating Alternative Branches of Plant Alkaloids”
DOI: 10.1038/s41467-022-28883-8
Christopher J. Vavricka *, Shunsuke Takahashi *, Naoki Watanabe, Musashi Takenaka, Mami Matsuda, Takanobu Yoshida, Ryo Suzuki, Hiromasa Kiyota, Jianyong Li, Hiromichi Minami, Jun Ishii, Kenji Tsuge, Michihiro Araki, Akihiko Kondo, Tomohisa Hasunuma (* equal contribution)
publishing magazines
Nature Communications

Related Links


Leave a Comment