🌐 Overview

Welcome to our cutting-edge computational pipeline designed to accelerate Alzheimer's Disease (AD) research. This project integrates advanced bioinformatics and cheminformatics, creating a seamless workflow from raw single-cell RNA sequencing (scRNA-seq) data to predictive Quantitative Structure-Activity Relationship (QSAR) modeling.

Our mission is to democratize access to powerful predictive tools, lowering the barrier to entry for researchers in the neurodegenerative disease space. This repository provides a comprehensive toolkit for data integration, cellular analysis, and machine learning-based bioactivity prediction.

You can access and use the live application at: https://QSARify.com

🔗 Project Workflow

✨ Features

This pipeline is organized into three core modules, each providing a distinct set of functionalities.

🔬 Single-Cell Analysis

• 🔗 Data Integration: Merges complex scRNA-seq datasets from multiple public studies (GSE138852, GSE157827, GSE175814, GSE163577) across various brain regions into a unified Seurat object.
• • Notably, the original studies encompassed over 500,000 cells, all of which were processed in our analysis. For ease of GitHub upload, a random subset of 25,000 cells (approximately 5,000 from each study) has been provided
• ✅ Quality Control & Normalization: Implements a robust QC pipeline that filters low-quality cells (e.g., mitochondrial content > 10%), normalizes data using SCTransform, and removes artifacts with DoubletFinder.
• 🧬 Cell Type Annotation: Automatically annotates cell clusters using the SingleR package with established reference datasets.
• 📊 Dimensional Reduction & Clustering: Leverages PCA for initial dimensionality reduction, Harmony for batch effect correction, and UMAP for visualization and clustering.
• 🔍 Differential Abundance: Employs MiloR to identify statistically significant changes in cell population abundance between experimental conditions.
• 📡 Cell-Cell Communication: Uses CellChat to infer and analyze intercellular communication networks, identifying key ligand-receptor interactions and signaling pathways.

🧪 QSAR Modeling & Bioactivity Prediction

• 🎯 Target Data Collection: Queries the ChEMBL database using UniProt IDs for key AD-related targets (MAO-B, COX-2, VISFATIN, BACE1, AChE). It retrieves bioactivity data (e.g., IC50 values) and calculates critical ADME properties (MW, LogP, HBD, HBA, Lipinski's Rule).
• 🧠 Machine Learning Pipeline:
  • Feature Engineering: Converts SMILES strings into 2048-bit Morgan fingerprints and calculates key physicochemical descriptors.
  • Imbalance Handling: Utilizes the SMOTETomek technique to address class imbalance in the bioactivity data.
  • Model Training & Tuning: Trains multiple baseline models (Random Forest, Gradient Boosting, Neural Network) and hyperparameter-tunes the top performer (Random Forest) to achieve an AUC of ~0.82 on the test set.
• 📈 Model Evaluation: Generates comprehensive visualizations for model performance, including ROC curves, accuracy plots, and F1-score comparisons.

🌐 Interactive Web API

• 🚀 RESTful Endpoints: A Flask-based API provides endpoints for health checks (/health), single predictions (/predict), and batch predictions (/predict_batch).
• 🎨 User-Friendly Interface: An intuitive web UI for interacting with the QSAR model.
  • Single & Batch Predictions: Supports bioactivity prediction for one or multiple compounds via SMILES input or file upload (.txt, .xls, .xlsx).
  • Dynamic Target Selection: Allows users to predict against a specific target or all available targets.
  • History Tracking: A sortable and filterable history tab keeps a record of all prediction tasks.
  • About Page: Contains project details and team information.

📂 Project Structure

AlzheimerDisease_FromSingleCell/
├── SingleCell/                # 🧬 scRNA-seq preprocessing (R)
│   ├── Merge_Data.R
│   └── SingleCell_Main.R
├── MiloR/                     # 🧬 Differential-abundance analysis (R)
│   └── MiloR_CellAbundance.R
├── CellChat/                  # 🧬 Cell–cell communication analysis (R)
│   └── CellChat.R
├── QSAR/                      # 🧠 QSAR modeling & web app (Python)
│   ├── figures/
│   │   └── roc_curves_comparison.png
│   ├── Data/
│   │   ├── chembl_results_P_27338_MAO-B_IC50_classified.csv
│   │   ├── chembl_results_P_35354_COX2_IC50_classified.csv
│   │   ├── chembl_results_P_43490_VISFATIN_IC50_classified.csv
│   │   ├── chembl_results_P_56817_BACE1_IC50_classified.csv
│   │   └── chembl_results_Q_04844_ACHE_IC50_classified.csv
│   ├── Model/
│   │   └── final_tuned_model.pkl
│   ├── templates/
│   │   └── index.html
│   ├── Target_Collection.ipynb
│   ├── Ligand_Final.ipynb
│   ├── app.py
│   └── requirements.txt
├── Data/                      # A large-scale analysis of over 500,000 cells was performed. A 25,000-cell subset (5,000 from each study) is provided on GitHub for convenience.
│   └── 25K_Sample.rds
└── README.md

📝 Detailed Script Information

Script 🖥️	Purpose 🎯	Key Libraries 🛠️	Output 📄
Merge_Data.R	Integrates raw scRNA-seq count matrices from multiple GSE studies.	Seurat, batchelor, SingleCellExperiment	A unified Seurat object containing all datasets.
SingleCell_Main.R	Performs QC, normalization, clustering, and cell type annotation.	Seurat, harmony, DoubletFinder, SingleR	A processed Seurat object with UMAPs and cell annotations.
MiloR_CellAbundance.R	Conducts differential abundance testing on cell neighborhoods.	miloR, SingleCellExperiment, ggplot2	Differential abundance statistics and visualizations.
CellChat.R	Infers and analyzes cell-cell communication pathways.	CellChat, Seurat, dplyr	Communication network data and plots (bubble plots, heatmaps).
Target_Collection.ipynb	Retrieves and preprocesses bioactivity & ADME data from ChEMBL.	pandas, chembl_webresource_client, rdkit	A cleaned DataFrame and exploratory data visualizations.
Ligand_Final.ipynb	Trains, tunes, and evaluates the QSAR machine learning model.	scikit-learn, imbalanced-learn, rdkit, pandas	A serialized model (.pkl) and performance plots.
app.py	Serves a Flask-based web API for on-demand bioactivity predictions.	flask, flask-cors, joblib, rdkit	JSON responses with predictions and confidence scores.
index.html	Provides the interactive front-end UI for the QSAR prediction tool.	HTML, CSS, JavaScript	An interactive web interface rendered in the browser.

🛠️ Installation

To set up the project environment, please follow these steps.

Step 1: Install R Dependencies 📦

(Required for SingleCell, MiloR, and CellChat analysis)

# Install core packages from CRAN
install.packages(c("Seurat", "dplyr", "ggplot2", "patchwork", "scater", "scran", "harmony", "batchelor", "SingleR"))

# Install Bioconductor packages
if (!requireNamespace("BiocManager", quietly = TRUE))
    install.packages("BiocManager")
BiocManager::install(c("SingleCellExperiment", "miloR", "glmGamPoi"))

Step 2: Install Python Dependencies 🐍

(Required for QSAR modeling and the Flask API)

# Clone the repository
git clone [https://github.com/xhammady/AD-scRNA2QSAR.git](https://github.com/xhammady/AD-scRNA2QSAR.git)
cd AD-scRNA2QSAR

# Install Python packages from requirements.txt
pip install -r QSAR/requirements.txt

Note: Key Python libraries include chembl-webresource-client, rdkit, scikit-learn, imbalanced-learn, pandas, flask, and flask-cors.

🚀 Usage

Follow this sequence to run the full analysis pipeline.

➡️ Step 1: Run the Single-Cell Bioinformatics Pipeline

1. 🧬 Merge Data: Execute SingleCell/Merge_Data.R to combine the raw count matrices.
2. 🔬 Preprocess & Cluster: Run SingleCell/SingleCell_Main.R to perform QC, normalization, integration, and annotation.
3. 🔍 Analyze Differential Abundance: Use MiloR/MiloR_CellAbundance.R to compare cell populations.
4. 📡 Infer Communication: Run CellChat/CellChat.R to analyze signaling pathways.

➡️ Step 2: Run the QSAR Cheminformatics Pipeline

1. 🧪 Collect Target Data: Open and run the QSAR/Target_Collection.ipynb notebook to query ChEMBL and generate the analysis dataset.
2. 🧠 Train ML Model: Open and run QSAR/Ligand_Final.ipynb to preprocess features, train the Random Forest model, and save the final .pkl file.

➡️ Step 3: Launch the Predictive API & User Interface

1. 🌐 Start the Server: From the command line, run the Flask application:

   python QSAR/app.py

2. 🎨 Access the UI: Open your web browser and navigate to http://localhost:5000/ or visit the live application at https://QSARify.com. You can now:
   • Enter a SMILES string for a single compound prediction.
   • Upload a file (.txt, .xls, .xlsx) for batch predictions.
   • View and manage results in the History tab.

🤝 Contributing

We welcome contributions to improve this project! Please fork the repository, create a new branch for your feature, and submit a pull request with a detailed description of your changes. Ensure you follow existing coding standards and include tests where applicable.

📜 License

This project is licensed under the MIT License. See the LICENSE file for more details.

🙏 Acknowledgments & Contributors

Project Team

• Amr Mohamed ElHefnawy
• Ibrahim Abdelkarim Hammad
• Mira Moheb Attia
• Abdelrahman Wagih
• Reem Sharaf EL-Deen Hassan
• Lorance Gergis Labeeb
• Prof. Dr. Sameh Ibrahim Hassanien

Tools & Data

• 🧬 Software & Libraries: Seurat, CellChat, MiloR, RDKit, scikit-learn, imbalanced-learn, Flask, DoubletFinder, SingleR, pandas, numpy, matplotlib.
• 🧪 Databases: We gratefully acknowledge the ChEMBL database for providing essential bioactivity data.
• 📊 Data Providers: This work would not be possible without the public datasets provided by Gene Expression Omnibus (GEO): GSE138852, GSE157827, GSE175814, GSE163577.

📧 Contact

• For questions or issues, please open an issue on GitHub or contact the project maintainers through mail.
• Amr Mohamed ElHefnawy
• Ibrahim Abdelkarim Hammad
• Mira Moheb Attia
• Abdelrahman Wagih
• Reem Sharaf EL-Deen Hassan
• Lorance Gergis Labeeb
• Prof. Dr. Sameh Ibrahim Hassanien