The Prometheus Large Language Model is a model that empowers individuals, regardless of their background or expertise, to harness the potential of synthetic biology and become beneficiaries of this cutting-edge science. We'll briefly introduce how Prometheus was designed and constructed in this page.

We first gathered relevant information about components from the iGEM database through web crawlers and official database interfaces. It utilizes fine-tuned ChatGLM3 and Prompt Engineering Llama3 to summarize the functional descriptions and usage contexts of these components based on the descriptions, organizing and storing them in a structured knowledge graph. Then, Prometheus uses the BioSentVec large language model to convert natural language information into embedding vectors for subsequent queries and matches.

Prometheus provides users with both Freshman Mode and Expert Mode interfaces. In Freshman Mode, users express their needs for target synthetic biology components or plasmids through natural language conversations with the Prompt Engineering-enhanced Llama 3 language model. Throughout this process, Prometheus continuously guides users to clarify their needs until an accurate, formatted list of component requirements is extracted. On the other side, for users with a certain foundation in synthetic biology, Prometheus offers Expert Mode, allowing direct input of component requirement lists to enhance communication efficiency.

After obtaining the component requirement list, Prometheus again uses BioSentVec to vectorize the natural language requirements in the list and matches them with the previously stored vectorized components in the iGEM database. The top 10 results with the highest matching degrees are then recommended by Llama 3, highlighting the components that best meet user needs; the remaining 9 components will also be output for user selection.

In addition, Prometheus automatically constructs complete plasmids that meet user needs based on the highlighted recommended components and identified usage contexts, facilitating users' subsequent experiments.

Prometheus also accepts user feedback, including ratings in the component recommendation list and a separate feedback webpage. We firmly believe that user feedback is a crucial factor in continuously improving model quality. In the future, Prometheus will fine-tune relevant large models based on user feedback data to enhance the accuracy of plasmid recommendations and the success rate of plasmid constructions.