"Prediction of Natural Product Biosynthesis Pathways... Expectation for Mass Production of Natural Product-Based Pharmaceuticals"

A model predicting the retro-synthetic pathways of natural products using deep learning has been proposed, providing a foundation for the mass production of natural product-based pharmaceuticals.

On the 16th, KAIST announced that Professor Sangkyu Kim’s research team from the Department of Biological Sciences and Professor Seongju Hwang’s research team from the AI Graduate School, led by Jaecheol Kim, developed a deep learning model predicting natural product biosynthetic pathways. In collaboration with Professor Jungbin Park’s research team at Pusan National University, they established an internet website to make this model accessible to everyone.

Plants produce complex natural products to respond to environmental stress. These natural products play an essential role in human survival as well. The fact that over 30% of FDA-approved small molecule drugs are based on plant natural products attests to this.

However, elucidating biosynthetic pathways is essential for utilizing and mass-producing natural products. On the other hand, the biosynthetic pathways of many medicinal natural products with complex structures remain unclear, so direct extraction from plants is commonly used.

For the same reason, research on biosynthetic pathways is challenging, but it has been believed that if these pathways can be elucidated and biosynthetic enzymes identified, the utilization value of natural products could be enhanced.

The joint research team started by tracing back the pathways of how plants synthesize substances (proposing retro-synthetic pathways) and succeeded in presenting a model that predicts the retro-synthetic pathways of natural products using deep learning.

In the study, the joint research team combined an advanced retro-synthesis model with biochemical intuition to successfully develop an artificial intelligence model that predicts natural product biosynthetic pathways.

The developed AI was named ‘READRetro,’ meaning ‘a model that reads retro-synthesis.’ This model was confirmed to demonstrate the best performance among AI models predicting natural product retro-synthesis, and it is significant that it was implemented (website established) for easy use by individual researchers.

Professor Sangkyu Kim said, “The research results are expected to be utilized from basic research understanding how plants produce complex natural products through various processes to synthetic biology research for mass production of natural product-based pharmaceuticals. The joint research team plans to continue research to predict enzymes mediating synthetic pathways or to improve the accuracy of retro-synthesis prediction for macromolecules.”

Meanwhile, this research was conducted with support from KAIST POST-AI, the National Research Foundation of Korea, and the Ministry of Science and ICT. Taein Kim, a combined master’s and doctoral student in the Department of Biological Sciences at KAIST, and Seul Lee, a combined master’s and doctoral student at KAIST AI Graduate School led by Jaecheol Kim, participated as co-first authors. The research results were published in the international academic journal New Phytologist.