BioNarrator-LLM

A Python-based framework that interprets real-time wet-lab sensor data into explainable biological insights using Retrieval-Augmented Generation (RAG), structured prompt engineering, rule-based validation, and explainability modules.

Based on the research paper: "BioNarrator-LLM: A Retrieval-Augmented Framework for Interpreting Real-Time Wet-Lab Sensor Data into Explainable Biological Insights" Rakshaiya Yadav G, Niranjana P Sathyabama Institute of Science and Technology, Chennai

---

What It Does

Takes raw wet-lab sensor readings (temperature, pH, microbial growth rate) and produces:

• A human-readable biological interpretation of the experimental state
• Likely cause of current conditions
• Recommended corrective action
• Rule-based validation with confidence scores
• Feature importance showing which parameter is driving the biological state

---

Sample Output

[INPUT] Temperature: 42°C | pH: 5.2 | Growth Rate: 0.3 OD/hr

BIOLOGICAL STATE SUMMARY
The organism is under severe combined stress. Elevated temperature (42°C)
is causing protein denaturation, while the acidic pH (5.2) is inducing
proton toxicity. This dual stress is reflected in the critically suppressed
growth rate of 0.3 OD/hr.

LIKELY CAUSE
Probable cause(s): excessive incubation temperature causing thermal
denaturation, acid accumulation possibly from fermentation byproducts,
nutrient depletion or toxic metabolite accumulation.

RECOMMENDED ACTION
  → Reduce incubation temperature to 35–37°C (currently 42°C)
  → Add buffer solution to raise pH towards 6.5–7.5 (currently 5.2)
  → Check nutrient availability and replenish growth medium

VALIDATION & CONFIDENCE
  • ALERT: Multiple stress conditions detected simultaneously.
  Confidence Score : 0.9
  Reliability      : High

FEATURE IMPORTANCE
  Temperature     █████ 27.9%
  pH Level        ██████ 31.9%
  Growth Rate     ████████ 40.2%

---

System Architecture

The framework consists of 5 core modules:

Module	Function
Data Preprocessing	Normalizes and classifies raw sensor values
Retrieval Module	Fetches relevant biological knowledge from knowledge base
Reasoning Engine	Generates structured biological narrative
Validation Layer	Rule-based checks with confidence scoring
Explainability Module	Feature importance and reliability assessment

---

Tech Stack

• Python — core language, no external dependencies
• Rule-based RAG — domain knowledge retrieval simulation
• Structured prompt engineering — converts sensor data to biological context
• Statistical validation — confidence scoring and anomaly flagging

---

Setup & Run

1. Clone the repository:

git clone https://github.com/Rakshaiya/Bionarrator-llm.git
cd Bionarrator-llm

2. No installations needed — pure Python

3. Run the framework:

python bionarrator_llm_local.py

---

Input Parameters

Parameter	Range	Unit
Temperature	20 – 45	°C
pH Level	4.5 – 8.5	—
Growth Rate	0.1 – 1.5	OD/hr

---

Use Cases

• Automated wet-lab monitoring
• Research analytics and data interpretation
• Smart healthcare and biomedical systems
• Educational tool for computational biology

---

Research Paper

This project is a working implementation of the BioNarrator-LLM framework proposed in our research paper, which achieved:

• 99.93% accuracy on simulated wet-lab dataset
• BLEU Score: 0.91 | ROUGE Score: 0.94
• Outperformed rule-based (85.4%) and statistical models (91.75%)

---

Author

Rakshaiya Yadav G

• GitHub: @Rakshaiya
• Email: rakshaiya115@gmail.com