Software

Modelling is one of synthetic biology's most powerful tools, allowing researchers to test designs, accelerate discovery, and refine experiments with greater confidence. Yet, for many students and iGEM teams, its potential remains locked behind a steep learning curve. Existing tools often require deep expertise in specialized formats like SBML (Systems Biology Markup Language), SBOL (Synthetic Biology Open Language), and many more, which can be difficult to create or modify without prior experience.

To bridge this gap, we developed BioMap: an intuitive platform that makes biological modeling accessible to everyone, regardless of their computational background.

BioMap eliminates technical barriers by leveraging natural language processing. Users can simply describe their biological system in plain English, and our integrated AI assistant converts it into a valid, simulatable model. For those who already have a model, BioMap seamlessly handles SBML file uploads. The platform then runs the simulation and presents the results through interactive graphs and clear visual summaries.

By simplifying the design, build, test, and learn cycle, BioMap empowers a new generation of scientists to harness the full power of modeling. Our complete project, including all documentation, is open-source and available on our team's GitHub for community use and future development.

We structured the development of BioMap using the Agile Scrum framework, allowing us to build, release, and adapt features incrementally. Our primary stakeholders were our Wet Lab team members, who needed to quickly prototype models without getting bogged down in low-level syntax.

To guide the project, we defined a series of Minimum Viable Products (MVPs). This approach allowed us to build a functional core first and then layer on advanced features, ensuring BioMap was always a working tool throughout its development.

This iterative roadmap ensured that BioMap was always functional while steadily adding the intelligence and interactivity that now make it a unique and powerful tool. A more detailed summary of the planning and development of the software tool can be found on the Engineering Cycle page.

BioMap is built on a modular, full-stack architecture designed for scalability and a seamless user workflow. The system is composed of several key components that work in concert:

• PostgreSQL Database: This relational database serves as the persistent storage layer for the application. It stores all user account information, projects, chat histories, and the finalized SBML models, ensuring data integrity and allowing users to save and resume their work across sessions.


Figure 1: BioMap system architecture showing the interaction between frontend, backend, database, AI model, and simulation engine.

• Python FastAPI Server: The backend acts as the central nervous system of the application, providing a RESTful API for all client interactions. It manages user authentication, handles all data operations with the PostgreSQL database, and orchestrates communication between the frontend client, the AI model, and the simulation engine.
• AI Language Model (OpenAI): This component is the core of our natural language modeling interface. The FastAPI server sends user prompts and the current model state to the AI, which interprets the biological descriptions and returns a structured JSON object representing the updated model. This process makes complex systems biology modeling accessible without requiring knowledge of SBML.
• Simulation Core (Tellurium & RoadRunner): This is the high-performance computational engine. The backend passes finalized SBML models to this core, which uses the libRoadRunner library to execute rapid time-course simulations. It returns the raw numerical data, which is then sent to the frontend for visualization.
• React Web Client: The frontend provides a dynamic and responsive user interface for the entire platform. Built with React and Vite, it allows users to manage projects, interact with the AI through a chat interface to build models, and visualize simulation results using Plotly.js for interactive charting. All communication with the backend is handled through asynchronous REST API calls.


Figure 2: BioMap's React-based user interface showing the model building and simulation workflow.

This modular architecture ensures that each part of the application can be developed, scaled, and maintained independently. The backend and frontend are containerized and deployed on a cloud platform, which creates a consistent and reliable environment, streamlines updates, and ensures the application is robust and efficient.

The implementation of BioMap is centered on a robust set of API endpoints that connect our user interface with the powerful backend services for model building and simulation. The user interface is broken down into three functional sections: the Project Dashboard, the Model Building Interface, and the Simulation & Analysis view.

Project Dashboard

Upon logging in, the user is presented with their project dashboard. This view is populated by making a GET request to the /projects/{user_id} endpoint, which retrieves all projects associated with their account. Users can initiate a new modeling workflow by clicking "New Project," which triggers a POST request to /projects/create. This action also automatically creates an initial chat session, seamlessly navigating the user to the Model Building Interface to begin their work.

Model Building (Chat Interface)

This is the core of BioMap, where natural language is translated into a formal systems biology model. The interface supports two primary methods for model creation:

1. Conversational Input: The user describes their biological system in plain English. Each message is sent via a POST request to the /chat/ endpoint. The backend then packages this message with the current JSON representation of the model and sends it to the OpenAI GPT-4o service. The AI's response, a JSON object containing a conversational reply and the updated model state (new_model_state), is parsed and displayed. A dedicated React component renders the new_model_state in real-time, providing the user with a continuously updated, structured view of their species, reactions, and events.
2. SBML Upload: Users can import existing models by uploading an SBML file (.xml or .sbml), which sends the file's content to the POST /chat/upload_sbml endpoint. The backend then parses this file to establish the initial model state for the session.

The sidebar in this view allows for full management of chat sessions within a project, using POST, PUT, and DELETE requests to endpoints like /chat/create and /chat/{chat_id} to handle creation, renaming, and deletion.

Simulation & Analysis

When a user is ready to simulate, they navigate to the "Simulation" tab. This action triggers a POST request to /model/finalize/{session_id}, where the backend converts the session's final JSON model into a valid SBML string. The frontend then calls the POST /simulate endpoint, sending the model_id and simulation settings (e.g., start/end time).

The results are rendered in an interactive graph using Plotly.js. A key feature of this view is the real-time parameter tuning. A control panel displays sliders for every parameter in the model. Adjusting a slider instantly triggers a new call to the /simulate endpoint with the updated parameter value, allowing users to perform rapid "what-if" analysis and observe how the system's dynamics change.

The "Report Summary" section integrates a rich text editor with our AI. Users can write their own observations or click "Generate" (POST /model/{model_id}/regenerate-summary) to have the AI produce a structured, scientific summary of the simulation results. For collaboration and documentation, users can export the model as an SBML file (.xml), the time-series data as a CSV file, or the plot as a PNG image.

Automated Model Refinement

In addition to the user-facing features, we have implemented an automated process to enhance model robustness. Before a model is finalized for simulation, the backend analyzes the reaction network to intelligently identify species that act only as sources (only ever produced) or sinks (only ever consumed). It then automatically sets the boundaryCondition attribute for these species to true. This automated step prevents common simulation errors related to mass balance and improves the validity of the model without requiring expert intervention from the user.

BioMap is a full-stack web application. You can access the live, deployed version directly in your browser or run a local instance for development.

Our development process followed an agile SCRUM methodology, which was instrumental in successfully delivering a functional and deployed application. This approach allowed our team to consistently iterate and ensure the core features of BioMap—translating natural language into simulatable SBML models—were robust and user-friendly.

While the primary goals were achieved, the project also presented significant challenges that led us to strategically defer two ambitious features.

These challenges provided valuable insights into the complexities of automated parameterization and dynamic biological visualization. The concepts remain compelling, and the groundwork laid during this project has created a solid foundation for pursuing them in future iterations of BioMap.

In under a year, our team successfully moved BioMap from concept to a fully deployed application, achieving our primary goal of making computational modeling accessible. While the core functionalities were successfully implemented, the project also presented significant technical challenges that led us to strategically defer two of our more ambitious features: automated parameter inference from databases and real-time SBOL visualization.

Despite these challenges, BioMap stands as a powerful tool that directly addresses a common bottleneck for iGEM teams: the high barrier to entry for computational modeling. By providing an intuitive platform where users can design and test systems in silico using natural language, our software allows teams to iterate on their designs and identify potential flaws before committing to expensive and time-consuming wet lab experiments. This workflow directly supports the "Design-Build-Test-Learn" cycle that is central to synthetic biology.

The groundwork laid in tackling the project's challenges has directly informed our path forward. We plan to revisit both automated parameter inference and dynamic visualization, building on the robust foundation we have now established. By continuing to lower the technical barriers, BioMap will empower future iGEM teams and researchers to innovate faster, making sophisticated modeling a standard and accessible part of the synthetic biology workflow.

AI: Artificial Intelligence

GPT: Generated pre-trained transformer

SBML: Systems Biology Markup Language

SBOL: Synthetic Biology Open Language

Agile: A methodology that focuses on continuous delivery and improvement

SCRUM: A type of Agile process that allows teams to get multiple small parts done at a time, called sprints, with continuous testing and feedback.

Minimum Viable Products (MVPs): Functional prototypes which represent different iterations of the software tool.

Tellurium: A comprehensive Python accessible SBML-compliant modelling platform for biochemical reaction systems.