Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

BioModelsRAG: A Biological Modeling Assistant Using RAG (Retrieval Augmented Generation)

Bhavyahshree Navaneetha Krishnan, Adel Heydarabadipour, Herbert Sauro

TL;DR

BioModelsRAG tackles the challenge of interpreting complex SBML biomodels by employing a Retrieval Augmented Generation workflow. It converts SBML to Antimony, segments and summarizes, stores the results in a ChromaDB vector store, and uses a cosine-similarity search to feed the LLM with relevant context. The system is deployed as a web app and CLI, enabling researchers to query specific biomodels in natural language with reduced hallucinations. This work demonstrates the viability of RAG for domain-specific biological data and points to future enhancements such as GraphRAG.

Abstract

The BioModels database is one of the premier databases for computational models in systems biology. The database contains over 1000 curated models and an even larger number of non-curated models. All the models are stored in the machine-readable format, SBML. Although SBML can be translated into the human readable Antimony format, analyzing the models can still be time consuming. In order to bridge this gap, a LLM (large language model) assistant was created to analyze the BioModels and allow interaction between the user and the model using natural language. By doing so, a user can easily and rapidly extract the salient points in a given model. Our analysis workflow involved 'chunking' BioModels and converting them to plain text using llama3, and then embedding them in a ChromaDB database. The user-provided query was also embedded, and a similarity search was performed between the query and the BioModels in ChromaDB to extract the most relevant BioModels. The BioModels were then used as context to create the most accurate output in the chat between the user and the LLM. This approach greatly minimized the chance of hallucination and kept the LLM focused on the problem at hand.

BioModelsRAG: A Biological Modeling Assistant Using RAG (Retrieval Augmented Generation)

TL;DR

BioModelsRAG tackles the challenge of interpreting complex SBML biomodels by employing a Retrieval Augmented Generation workflow. It converts SBML to Antimony, segments and summarizes, stores the results in a ChromaDB vector store, and uses a cosine-similarity search to feed the LLM with relevant context. The system is deployed as a web app and CLI, enabling researchers to query specific biomodels in natural language with reduced hallucinations. This work demonstrates the viability of RAG for domain-specific biological data and points to future enhancements such as GraphRAG.

Abstract

The BioModels database is one of the premier databases for computational models in systems biology. The database contains over 1000 curated models and an even larger number of non-curated models. All the models are stored in the machine-readable format, SBML. Although SBML can be translated into the human readable Antimony format, analyzing the models can still be time consuming. In order to bridge this gap, a LLM (large language model) assistant was created to analyze the BioModels and allow interaction between the user and the model using natural language. By doing so, a user can easily and rapidly extract the salient points in a given model. Our analysis workflow involved 'chunking' BioModels and converting them to plain text using llama3, and then embedding them in a ChromaDB database. The user-provided query was also embedded, and a similarity search was performed between the query and the BioModels in ChromaDB to extract the most relevant BioModels. The BioModels were then used as context to create the most accurate output in the chat between the user and the LLM. This approach greatly minimized the chance of hallucination and kept the LLM focused on the problem at hand.
Paper Structure (45 sections, 5 equations, 3 figures)

This paper contains 45 sections, 5 equations, 3 figures.

Table of Contents

  1. Introduction
  2. Methods
  3. Downloading, converting, and segmenting BioModels
  4. Conversion of the Antimony model into plain text
  5. Storage of the Antimony models in a vector database
  6. Receiving user's query
  7. Database querying
  8. Creation of prompt for final response
  9. Generation of final response
  10. Results
  11. Implementation/Deployment
  12. Selection of specific BioModels
  13. Conversation interface
  14. Conclusions and Implications
  15. Documentation
  16. ...and 30 more sections

Figures (3)

  • Figure 1: A visualization of the pipeline with the exact steps labeled corresponding to the section name
  • Figure 2: UI with dropdown for choosing specific biomodels to analyze. The user can enter a search query and then choosing multiple or single models from the dropdown. The dropdown will contain all of the models that matched with the query.
  • Figure 3: The UI for the chat portion is shown above. The user's query will be displayed with a grey background, and the LLM's output will be displayed right below it. A textbox will also be right below the LLM output so that the user can continue entering queries.