🤖🧬 Phenotype RAG: Bio-Phenotype Insights Assistant

bio-phenotype.mp4

📌 You can explore and interact with the Bio-Phenotype by accessing the app through the following link: https://dry-recipe-9383.ploomberapp.io.

🧬 Project Overview

This project, Phenotype RAG, was developed as the final assignment for the LLM Zoomcamp. It implements a Retrieval-Augmented Generation (RAG) system that intelligently answers questions related to phenotypes by utilizing both a knowledge base and large language models (LLMs). The system is designed to assist with queries about phenotypes in fields such as genetics, evolutionary biology, and medical diagnostics. By integrating retrieval and generation capabilities, the project provides precise and contextually accurate information, making it a powerful tool for phenotype-related research and clinical applications.

🧬 Problem Description

Phenotyping plays a crucial role in various domains like genetics, evolutionary biology, and medical diagnostics, helping researchers and clinicians understand the observable traits influenced by genetic and environmental factors. However, the complexity and vastness of phenotype data make it challenging to access and retrieve relevant information efficiently. This project addresses the challenge by developing an intelligent assistant capable of answering complex phenotype-related queries. By leveraging RAG techniques, the system combines the reasoning ability of LLMs with the precision of a curated knowledge base, making the retrieval of phenotype information more accessible and accurate for researchers, healthcare professionals, and educators.

🧬 Project Objectives

The Phenotype RAG project aims to achieve the following objectives:

1. Enhance Data Retrieval: Implement a Retrieval-Augmented Generation (RAG) system to efficiently access and retrieve accurate information about phenotypes from a comprehensive knowledge base.
2. Improve Query Accuracy: Utilize advanced language models to reformulate and optimize queries, ensuring that the answers provided are contextually relevant and precise.
3. Offer Educational Value: Create an accessible platform for students and professionals to learn about phenotyping, improving their grasp of complex concepts through a user-friendly interface.
4. Ensure Scalability and Flexibility: Develop a system with a flexible architecture that can integrate with various tools and adapt to different research needs, promoting scalability and adaptability in diverse applications.
5. Foster Collaboration: Make the project's code and documentation available to the community, encouraging collaborative development and knowledge sharing to advance the field.

🧬 Technologies and Tools Used

⚗️ Key Technologies

Anaconda: Used for managing dependencies and environment configurations.
Docker: Containerizes the application for easy deployment and consistent execution across different platforms.
Grafana: Provides monitoring and visualization dashboards to track application performance and usage metrics.
Streamlit: Offers a user-friendly interface for interacting with the Phenotype RAG system.
Prefect: Orchestrates data ingestion workflows to ensure smooth and automated processes.

🧬 LLMs Used

gemma2-9b-it: Utilized for question reformulation, optimizing queries for better understanding.
mixtral-8x7b-32768: Powers the retrieval-augmented generation by processing large volumes of text and delivering more contextually accurate answers.
all-MiniLM-L6-v2: Handles embedding generation and semantic search, allowing for precise query-to-answer matching.
Groq: Integrates with the system for efficient vector processing during the search phase.
Pinecone: Manages vector indexing and provides fast, scalable retrieval of information using semantic search.

⚗️ Other Tools Used for Development

Pytest: Ensures code reliability through unit and integration tests.
Git: Version control for tracking changes and collaboration.
Visual Studio Code: Integrated development environment (IDE) for writing and debugging code.
Jupyter Notebook: Facilitates exploratory data analysis and preprocessing through interactive notebooks.
PostgreSQL: Relational database used for storing and querying structured data.

🧬 Project Structure

The project is organized into the following directories and files:

phenotype-rag/
├── bio-phenotype/                        # Root folder for the main application logic
│   ├── data/                             # Directory to hold project-specific datasets 
│   │   └── bio-phenotype.csv             # Main dataset: includes phenotype-related questions and answers
│   ├── sql/                              # Directory for database management and schema scripts
│   │   ├── .env                          # Environment file storing sensitive credentials database connection strings
│   │   └── create_table.py               # Python script to automate the creation of tables in PostgreSQL
│   ├── tests/                            # Directory for unit tests to ensure code quality and correctness
│   │   └── test.py                       # Python script containing test cases for core functionalities of the project
│   ├── __init__.py                       # Initializes the `bio-phenotype` package, making its modules importable across the project
│   ├── main.py                           # Streamlit application entry point; defines the UI and handles user interaction
│   ├── prefect_ingest.py                 # Prefect workflow script that automates data ingestion and processing tasks
│   ├── requirements.txt                  # Lists Python dependencies needed to run the project (for pip-based installations)
│   └── utils.py                          # Contains utility functions for data processing, I/O operations, and common tasks
├── data/                                 # Contains raw data files that can be accessed across different components
│   └── bio-phenotype.csv                 # Same dataset as in `bio-phenotype/data`, accessible for testing and backup
├── grafana/                              # Directory for Grafana monitoring setup
│   └── monitoring/
│       ├── docker-compose.yaml           # Docker Compose configuration for setting up Grafana
│       └── grafana_datasources.yaml      # Configuration file that defines the data sources Grafana will connect to PostgreSQL
├── images/                               # Directory for storing project-related images and screenshots
│   ├── app.png                           # Screenshot of the Streamlit app's interface
│   ├── grafana.png                       # Screenshot of the Grafana monitoring dashboard, displaying key metrics
│   ├── gloq.png                          # Screenshot of Groq AI acceleration with integrated API Keys
│   └── pinecone.png                      # Screenshot of Pinecone vector database powering semantic and similarity searches
├── notebook/                             # Directory containing Jupyter notebooks for exploratory data analysis (EDA) and model experimentation
│   ├── .env                              # Environment file specifically for notebook-related configurations (API keys, credentials)
│   └── vector_Indexing_.ipynb            # Notebook for vectorizing data and indexing it into the semantic search system (Pinecone)
├── docker-compose.yaml                   # Primary Docker Compose file to orchestrate multi-container setups, including app, database, and Grafana
├── README.md                             # Project documentation with detailed instructions on usage, setup, and project purpose
├── requirements.txt                      # Python dependencies for the entire project (ensuring the environment is consistent across machines)
└── test.py                               # Standalone test script covering various components, including ingestion, database interactions, and the API

🧬 Phenotype Dataset

The dataset used for this project contains questions and answers about phenotypes, with a focus on genetic research, evolutionary biology, and medical diagnostics. It explores how phenotypic traits relate to cognitive function, disease susceptibility, and treatment outcomes, highlighting the role of phenotyping in personalized medicine. The dataset also covers the impact of traits on aging, chronic diseases, and mental health disorders. Phenotypic trait analysis is crucial in understanding genetic predispositions, environmental adaptations, and evolutionary processes. This resource supports the development of diagnostic tools, therapeutic strategies, and health interventions by linking observable traits to genetic and environmental factors. Additionally, it is valuable for research in agricultural phenotypes, such as plant growth and disease resistance.

📝 Some Questions and Answers

🧬 Project Execution Locally

⚗️ Pre-requisites

Ensure the following are installed on your machine:

Anaconda (latest version)
Python (version 3.10 or later)
PostgreSQL (latest version)
Grafana (latest version)

⚗️ Environment Setup

Clone the repository:

git clone https://github.com/nathadriele/biophenotype-rag.git
cd bio-phenotype

Create and activate the virtual environment:

conda create -n bio-phenotype python=3.10
conda activate bio-phenotype

Install dependencies:

pip install -r requirements.txt

🧬 Data Exploration and Preprocessing

Start the vector_Indexing_.ipynb notebook with Jupyter:

jupyter notebook

🧬 Running the Application

To run the application, you will need access keys (API Key) for both GroqCloud and Pinecone. You will create and substitute them, as well as create an Index in Pinecone. You will need accounts on both platforms.

Step 1: Create API Key on GroqCloud

Create or log into your GroqCloud account and navigate to API Keys > Create API Key.
Copy and save the Key in a text editor for later use.

Step 2: Create an Index on Pinecone

On the Pinecone website, go to Indexes > Create Index.
Configure the index as follows:
- Default / bio
- Dimensions: 384
- Metric: Cosine
- Capacity mode: Serverless
- Cloud provider: AWS
- Region: Virginia | us-east-1
Complete the setup by clicking on Create Index.

Note: The region can be changed without significantly affecting the code. However, altering other configurations would require significant code adjustments.

Step 3: Add the API Keys to Environment Files

After completing the previous steps, add your API keys to the .env files in the notebook and lang-bio-groq folders, as shown below:

Make sure to replace your-pinecone-api-key and your-groqcloud-api-key with the actual keys you generated earlier.

Step 4: Running the Application Locally

To run the application locally, you may need to adjust the configurations in the .env file to match your environment. This also applies to the Grafana setup parameters shown below.

In the Anaconda Prompt, ensure you are in the lang-bio-groq folder and run the following command:

streamlit run main.py

🧬 Monitoring and Performance Metrics

Grafana is used to monitor performance, and the image displays a dashboard configured with key performance metrics. In this example, it is evident:

Average Response Time: The current average response time, which is tracked in real-time to ensure system responsiveness.
Record Count by Month: This chart tracks the number of records entered into the system.
Total Conversations: The gauge shows a total of conversations monitored, with the status represented in green, indicating acceptable levels.
Distribution of Questions and Answers: The average question length, and the average response length is significantly higher, at 161 characters. This highlights the tendency for longer responses compared to the questions.

🧬 Contribution of the Phenotype RAG Application

The Phenotype RAG: Bio-Phenotype Insights Assistant enhances research and practice in genetics and medical diagnostics by integrating retrieval and generation of phenotype information. It facilitates efficient access to complex data, supports accurate diagnostics, and provides a valuable educational tool. With flexible architecture, the application improves interaction with large volumes of data and fosters innovation through a collaborative and accessible approach for the community.

More Information

This project was developed as the final assignment for the LLM Zoomcamp course.