team3 — Per-cell quantification of Pol II nuclear localisation

Hackathon project analysing confocal microscopy images of HEK293 cells expressing Avi-tagged RNA Polymerase II.

Main finding

Per-cell analysis of n=11 transfected cells across 6 imaging fields shows that overexpressed Avi-Pol II accumulates in the cytoplasm rather than the nucleus (local ratio 0.47 ± 0.18). The signal is real — 25× above the antibody background measured from the no-transfection control (Sample 2). This quantifies the known localisation issue and motivates adding a nuclear localisation signal (NLS) in future constructs.

Pipeline

.czi file → split 4 channels
  ├── DAPI  → Cellpose (deep learning) → Nuclear mask
  └── Avi   → Otsu threshold + morphology cleanup → Avi mask
                                              ↓
                                  Boolean intersection
                                              ↓
                              Transfected nuclei (cells of interest)
                                              ↓
                              Per-cell ratios:
                                - Global: nucleus / image-wide outside
                                - Local:  nucleus / same cell's cytoplasm

Key numbers (Sample 7, full system)

Metric	Value	Interpretation
Transfected cells	n=11 across 6 fields	Sufficient for statistics
Global ratio (nucleus / outside)	15.7× ± 8.9	Signal is real, 10× above antibody background
Local ratio (nucleus / cytoplasm)	0.47 ± 0.18	Pol II is twice as concentrated in cytoplasm
Sample 2 background	1.56×	Antibody-only control baseline
Pipeline false positives in 4 controls (S1, S2, S3, S5)	0 / 95 nuclei	Pipeline doesn't over-call transfection

Setup (Mac)

python3.11 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Place .czi files in data/ (not tracked by git).

How to use

Open notebooks/FINAL_pipeline.ipynb and run cells sequentially. The pipeline processes Sample 7 replicates (7_1.czi through 7_6.czi), the antibody control (Sample2.czi), and additional controls (Sample1, Sample3, Sample5).

Outputs are saved to results_v2/:

• FIGURE_1_pipeline_overview.png — full pipeline visualisation (methods slide)
• FIGURE_2_background_validation.png — Sample 2 control vs Sample 7 (validation slide)
• FIGURE_3_main_results.png — per-cell ratios for n=11 cells (main result slide)
• sample7_all_cells.csv — raw per-cell measurements

Repository structure

.
├── notebooks/
│   ├── FINAL_pipeline.ipynb         ← MAIN ANALYSIS (use this)
│   ├── 05_clean_pipeline.ipynb      (earlier per-sample version)
│   └── older notebooks              (development history)
├── results_v2/                       ← Final figures and tables
├── results/                          (earlier per-sample outputs)
├── data/                             (gitignored — raw .czi files)
├── requirements.txt
├── .gitignore
└── README.md

Tools used

• Cellpose (Stringer et al., Nature Methods 2021) — pre-trained U-Net for nucleus segmentation
• scikit-image — Otsu thresholding, morphological operations, region properties
• aicspylibczi — .czi file reader
• pandas, matplotlib — data tables and visualizations

Team3