🧬 Bacterial Single-Cell RNA-Seq Analysis (microSPLiT)

An end-to-end Bioinformatics and Machine Learning pipeline for processing, analyzing, and interpreting bacterial single-cell RNA-sequencing (scRNA-Seq) data. This project utilizes the microSPLiT protocol applied to Bacillus subtilis across various growth stages, demonstrating how unsupervised computational workflows can unveil hidden biological states and rare subpopulations.

---

🚀 Key Features & Methodologies

• Data Quality Control (QC): Filtering low-quality bacterial cells based on gene counts and total transcripts (Scanpy).
• Dimensionality Reduction: Extracting biologically relevant signals using Highly Variable Genes (HVGs), Principal Component Analysis (PCA), and Uniform Manifold Approximation and Projection (UMAP).
• Graph-Based Clustering: Unsupervised community detection via the Leiden Algorithm to partition transcriptomic profiles without experimental bias.
• Differential Expression (DE): Identifying cluster-specific biomarkers using the Wilcoxon rank-sum test to reveal active metabolic pathways and cell fates.

---

📁 Project Structure

The analysis is structured into 3 modular Jupyter Notebooks:

1. 01_quality_control.ipynb: Initial data ingestion, parsing metadata (OD stages), metadata feature engineering, and stringent single-cell QC filtering.
2. 02_normalization_and_dimred.ipynb: Data scaling, selection of top 1,500 highly variable genes, PCA execution, and building the neighborhood graph for UMAP mapping.
3. 03_clustering_and_markers.ipynb: Unsupervised cell clustering, differential expression analysis, and marker gene visualization.

---

📊 Key Results & Insights

1. Unsupervised Clustering vs. Growth Dynamics

By embedding ~19,000 cells into a low-dimensional UMAP space, the unsupervised Leiden algorithm successfully recaptured the continuous biological trajectory of bacterial growth stages, mapping precisely from early exponential phases (OD0.4/OD1.0) to late stationary phases (OD2.7/OD3.2).

2. Automated Marker Gene & Subpopulation Discovery

Using differential expression analysis, we extracted unique transcriptomic signatures for each cluster. The generated Dotplot demonstrates stark expression boundaries between metabolic states:

• Translation Machinery (Clusters 0 & 3): High upregulation of ribosomal proteins (fusA, rplC, rplV), marking active protein synthesis during exponential growth.
• Metabolic Transition (Cluster 2): Shift towards acetoin catabolism (acoA, acoC) as primary carbon sources deplete.
• Stress Response & Surfactin Production (Cluster 4): Activation of stationary phase resilience genes (srfAA, srfAB).
• Rare Biological Event (Cluster 7): Isolation of a rare subpopulation (~132 cells) undergoing prophage induction (activation of the dormant PBSX prophage via xkdK, xkdO, xkdG), showcasing the power of single-cell resolution over bulk RNA-Seq.

---

🛠️ Environment Setup & Installation

To reproduce this analysis locally, ensure you have miniconda or anaconda installed, then run:

Create and activate the conda environment

conda create -n bacterial_sc_env python=3.10 -y conda activate bacterial_sc_env

Install required bioinformatics and plotting libraries

pip install scanpy pandas numpy matplotlib igraph leidenalg h5p