BioGPT

BioGPT is an API that makes proteomic data exploration simple and intuitive through natural language queries. Built on the UniProt biological database, it allows researchers, students, and developers to access complex biological information without needing SQL expertise or specialized file handling.

By combining semantic search and structured querying, BioGPT connects everyday language with biological data, leveraging modern AI models to interpret questions and deliver human‑readable answers.

Key Features

• Natural Language Interface: Ask biological questions in plain text;
• Hybrid AI Engine: Dynamically choose between semantic search (RAG) for conceptual questions or precise filtering (Text‑to‑SQL) for data extraction;
• Proteomic Focus: Optimized for protein‑related queries using UniProt data;
• RESTful API: Easy integration into bioinformatics pipelines, apps, or research tools.

Core Technologies

Layer / Component	Technology	Role
Hybrid Logic	RAG (Retrieval-Augmented Generation)	Handles conceptual queries; interprets semantic meaning and synthesizes narrative answers.
Structured Query	Text-to-SQL (Rule-Based)	Translates natural language into executable SQLite queries for precise filtering and counting.
NLP Models	Sentence Transformers (MiniLM)	Generates embeddings for semantic search (RAG retrieval step).
NLP Models	GPT-2 (Open Source)	Synthesizes coherent, human-readable answers from retrieved biological context (RAG generation step).
Data Infrastructure	Pandas & SQLite	Loads UniProt TSV data into an in-memory SQLite database for fast, thread-safe structured querying.
Web Framework	Flask (Python)	Provides the RESTful API (/api/biogpt), handling input validation and JSON output.

Engines

• RAG Engine: Best for semantic queries and synthesized answers.
  • Example: "Describe the function of this protein."
• Text-to-SQL Engine: Best for precise filtering, counting, and numerical comparisons.
  • Example: "List proteins with mass > 100000."

Installation & Setup

1. Clone repository

2. Install Dependencies

pip install pandas flask sentence-transformers scikit-learn transformers torch accelerate

3. Run Application

Important: Execute the application from the project root directory using the module flag (-m) to ensure correct package imports.

# From the BioGPT/ root directory:
python3 -m src.app

The API will start at http://127.0.0.1:5000/

API Usage

The BioGPT API exposes a single endpoint for all query processing.

Endpoint

Method	URL	Description
POST	/api/biogpt	Submits a natural language query to the hybrid engine.

Request Body

Field	Type	Description	Required	Options
query	string	Natural language question.	Yes	-
model_type	string	Selects processing engine. Defaults to rag.	No	rag, sql

---

Example 1: rag (Semantic Search)

Request:

curl -X POST http://127.0.0.1:5000/api/biogpt \
     -H "Content-Type: application/json" \
     -d '{"query": "What is the role of Complement C1q in the immune system?", "model_type": "rag"}'

Response (JSON):

{
   "model_type":"RAG",
   "response":"What is the role of Complement C1q in the immune system?\n\nWhat is the role of Complement C1q in the immune system?\n\nWhat is the role of Complement C1q in the immune system?\n\nWhat is the role of Complement C1q in the immune system?\n\nWhat is the role of Complement C1q in the immune system?\n\nWhat is the role of Complement C1q in the immune system?\n\nWhat is the role of Complement C1q in the immune system?\n\nWhat is the role of Complement C1q in the immune system?\n\nWhat is the role of Complement C1q in the",
   "retrieved_contexts":[
      {
         "Context_Bio":"Protein: Glucan endo-1,3-beta-D-glucosidase 1 (Endo-1,3-beta-glucanase 1) (EC 3.2.1.39) (Laminarinase) (RmLam81A). Organism: Rhizomucor miehei. Subcellular Location: SUBCELLULAR LOCATION: Secreted, cell wall {ECO:0000250|UniProtKB:P53753}.. Biological Process: cell wall organization [GO:0071555]; polysaccharide catabolic process [GO:0000272]. Molecular Function: endo-1,3(4)-beta-glucanase activity [GO:0052861]; glucan endo-1,3-beta-D-glucosidase activity [GO:0042973]",
         "Entry_Name":"ENG1_RHIMI",
         "Similarity_Score":"0.7025"
      },
   ],
   "status":"success",
   "user_query":"What is the role of Complement C1q in the immune system?"
}

---

Example 2: sql (Text-to-SQL)

Request:

curl -X POST http://127.0.0.1:5000/api/biogpt      -H "Content-Type: application/json"      -d '{"query": "Count how many proteins are secreted.", "model_type": "sql"}'

Response (JSON):

{
   "count":33597,
   "data":[
      
   ],
   "message":"Query executed successfully, found 33597 records.",
   "model_type":"SQL",
   "sql_query":"SELECT COUNT(*) FROM proteins WHERE 1=1 AND Subcellular_location_CC LIKE '%Secreted%'",
   "status":"success",
   "user_query":"Count how many proteins are secreted."
}