Chinese researcher taps large language models to make TCM drug more innovative, productive

Dajing county of Wenzhou, Zhejiang province, is making an attempt to feed data of the traditional Chinese medicine shi hu, or dendrobium, into medical large language models.

A team of about five people started the pilot program in July. Song Bowen, a lecturer at the School of Medical Humanities and Management at Wenzhou Medical University, has led the program.

As Song explained, LLM models are more widely used in sectors such as automotive production. Its application in the medical field is still at a preliminary stage worldwide. Meanwhile, most medical LLMs now have their data centered on the Western clinical medicine. By training the model with TCM data, such as dendrobium, will make it more precise, said Song. 

With LLM's help, Song's team was more than surprised to find that Russian and Japanese professionals have applied a large number of dendrobium-related patents. This will be quite useful for dendrobium growers in Dajing who would like to tap into the overseas markets, he said. 

"Based on this model, we can look for the possible paths for product innovation, make in-depth analysis for the whole industry and even discover market uncertainties and risks," he said.

About 6,000 mu (4 square kilometers) of dendrobium is grown in Dajing, hiring more than 30,000 people - nearly one-third of the county's total population. The annual dendrobium industrial output has exceeded 1.5 billion yuan ($206 million) in this county.