Design
Before the project was launched, we couldn't determine which carbon source, red algae or glucose, would yield higher benefits for the synthesis of rare ginsenoside Rh1. To further explore the feasibility of the project, we interviewed Dr. Yan Xiaofang, who studies synthetic biology and intelligent protein design at the School of Biological Science and Engineering, South China University of Technology.
Build
Dr. Yan Xiaofang suggested from the perspective of bioengineering that we could construct a metabolic network of Saccharomyces cerevisiae through constraint modeling and analyze metabolic flux to select carbon sources.
Test
We have built an FBA (Flux Balance Analysis) platform, and based on simulation results, we have found that a mixed carbon source consisting of galactose and AHG (derived from red algal biomass) not only supports efficient Rh1 production, but also significantly increases the synthesis flux compared to the strategy of using only glucose.
Learn
According to the simulation results of FBA, red algae can be used as a new carbon source to replace traditional glucose carbon source and efficiently synthesize Rh1, which verifies the feasibility of the project.
CYCLE 1: The development of FBA (Flux Balance Analysis) software
Design
After initially verifying the feasibility of utilizing red algae, we plan to construct a yeast engineering strain capable of efficiently utilizing red algae. However, due to the wide range and variety of related hydrolytic enzyme gene sources, it is difficult to find suitable target genes from a vast amount of data. Therefore, we interviewed Dr. Liang Hui , who studies yeast metabolic engineering at the School of Food Science and Engineering, South China University of Technology. [7][8][9]
Build
Dr. Liang Hui provided us with some literature she had used in her research, including many hydrolytic enzyme genes from red algae, and provided guidance for us to search for related genes in the NCBI database. [10][11][12]
Test
We reviewed these relevant literature and searched for related genes in the NCBI database. Finally, we screened out three different sources of agarase and two different sources of neogalactobiohydrolase. We constructed six plasmids and introduced them into yeast cells. Through the Lugol's iodine plate method, we qualitatively detected that all six strains had the ability to secrete hydrolytic enzymes
Learn
The experimental results show that the yeast engineering strain capable of utilizing red algae has been preliminarily constructed. We plan to further search for strains with higher hydrolytic enzyme activity.
CYCLE 2: Constructing recombinant vectors expressing agarase and neoagarobiose hydrolase
Design
After transforming the five genes for Rh1 synthesis into the strain Sq-0 using high-copy and low-copy plasmids, we conducted fermentation experiments using glucose. After ensuring the accuracy of all operations and gene sequencing, and after multiple repeated experiments, we did not detect any rare ginsenoside Rh1 or any intermediate products. To solve the problem of being unable to produce rare ginsenoside Rh1, we interviewed Professor Wu Hong from the School of Food Science and Engineering at South China University of Technology in the field of food biotechnology, and Professor Chen Gu from the field of microbial functional genomics and nutritionomics.
Build
After communicating with the two professors and consulting numerous materials, we still failed to find the problem of experimental failure. However, Professor Chen Gu suggested that we change the method and integrate the five genes for Rh1 synthesis into the genome of yeast cells before conducting the experiment again. [13][14][15][16]
Test
After using CRISPR-Cas9 technology to integrate five exogenous genes into the corresponding sites of the yeast cell genome in sequence, we successfully obtained all intermediate products and Rh1 through fermentation experiments.
Learn
Our experimental scheme has been successfully demonstrated to be feasible, achieving the de novo synthesis of rare ginsenoside Rh1 by Saccharomyces cerevisiae using red algae. In the future, we plan to further explore the reasons for the failure of transforming high and low copy number plasmids into yeast for the production of rare ginsenoside Rh1.
CYCLE 3: Constructing engineered strains capable of producing rare ginsenoside Rh1 using high and low copy number plasmids
Design
In our research on launching the sub-combination prediction model, we faced the problem of balancing the number of hypotheses, variable dimensions, and computational accuracy. To address this issue, we interviewed Professor Liu Shenquan from the School of Mathematics at South China University of Technology, who specializes in applied mathematical modeling.
Build
Professor Liu Shenquan affirmed the feasibility of the project in experimental design, dry experiment modeling, and practice of rare ginsenoside biosynthesis. He provided directional guidance for core issues: establishing chemical reaction kinetics equations, analyzing the impact of conversion reaction processes and iterative parameter changes on product yield to achieve a balance between the number of assumptions and computational accuracy. By using a limited range and exhaustive combination calculation method, the robustness of the correction was ensured.
Test
We constructed relevant ordinary differential equations based on the guidance and chemical reaction kinetics model, and limited the initial parameter range through reviewing papers and experiments. We combined traversal calculations to achieve parameter recalibration after adding new promoters, and optimized the promoter combination prediction model.
Learn
This guidance clarifies the direction of experimental modeling optimization, addresses the three major challenges of the model, and provides mathematical theoretical analysis support for subsequent launch of sub-combination precision prediction.
CYCLE 4: Optimization and parameter correction scheme for promoter combination prediction model
Design
To address the issue of insufficient professionalism of agents in software and occasional "illusion" problems, we specially invited Dr. Zhang Tong from the School of Computer Science and Engineering, South China University of Technology, who specializes in AI and affective computing, for an interview. Dr. Zhang pointed out that model deployment should focus on the application of knowledge distillation technology, and suggested building small-scale domain models tailored to specific projects, emphasizing deep customization and optimization rather than simply calling generic models. In addition, he particularly emphasized the key role of expert experience in model application, proposing that a systematic construction of relevant knowledge bases should be carried out to enhance the professionalism and reliability of models.
Build
Based on expert advice and comprehensive team analysis, we have decided to start with knowledge base construction and Prompt engineering to systematically enhance the professional capabilities of the Agent and suppress hallucination phenomena. In terms of technology selection, we have adopted DeepSeek-R1-Distill-Qwen-7B as the basic language model, integrated feedback from team members and relevant scientific researchers, as well as biological domain expertise, to build a knowledge base specifically serving functional modules such as FBA principle explanation and operation assistance. Based on this, we have optimized and designed more structured and logically rigorous Prompt templates.
Test
After system testing and evaluation, the agent has shown significant improvement in answering user questions: professional explanations are more accurate, irrelevant or incorrect content is significantly reduced, and the logic and consistency of answers have been enhanced.
Learn
Significant progress has been made in this stage of development. The AI Agent has not only improved professional quality in responding to user needs, but also significantly improved user experience, effectively alleviating issues such as confusing answers and frequent hallucinations. Looking ahead, we plan to further develop attention-enhancing training based on existing models, improve contextual memory mechanisms, and continuously optimize model weight configurations to better adapt to diverse and highly complex interaction scenarios.
CYCLE 5: FBA Graphical Interactive Software - Enhancing Agent Professionalism and Reducing Illusions
Design
The product faces a key market positioning issue: which efficacy should be promoted as the core selling point to most effectively impress target consumers?
Candidate efficacy claims include:
1. Powerful anti-wrinkle and firming
2. Extreme repair and stabilization
3. Excellent brightening and yellowing removal
Different efficacy positioning will determine our product formula fine-tuning, visual design, marketing content and channel selection.
Build
To address this issue, we consulted Professor Liu Wei from the College of Life Sciences at South China Agricultural University, who specializes in the study of the application of natural products in skin care mechanisms. Professor Liu analyzed and pointed out: "The experimental data of Rh1 shows that it has potential in anti-wrinkle, repair, and whitening, but the market does not need a 'one-size-fits-all solution'. Consumers seek 'specialized and refined solutions' in high-end functional skin care products. I suggest that you go beyond traditional questionnaire research and adopt 'concept-formula integration testing' to measure the full-chain data of 'attention-consultation-information retention-transformation'. Real behavioral data is far more persuasive than verbal claims of preference."
Test
Based on Professor Liu's advice, we conducted market research in three areas: material preparation, channel selection, and data tracking. Based on key metrics, we verified the core market pain point: "strong anti-wrinkle and firming" is the most urgent and willing-to-pay demand of current high-end skin care consumers.
Learn
Based on data feedback, we will launch the main repair series first, while using "anti-wrinkle" as a technical reserve for future series, optimizing product mix and research and development routes.
CYCLE 6: Business Plan