docs/REPRODUCTION_GUIDE.md

# Reproduction Guide

Step-by-step instructions to reproduce all analyses from:

> Harrizi S., Nait Irahal I., Kabine M. *Forbidden Words: How Nullomer Constraints Channel Regulatory Evolution in ABC Transporters.* (2025)

---

## Prerequisites

| Requirement | Minimum |
|---|---|
| Python | 3.8 or higher |
| RAM | 8 GB (16 GB recommended) |
| Disk space | ~5 GB (genome + results) |
| Internet | Required (genomes downloaded at runtime) |

---

## Setup

```bash
# Clone the repository
git clone https://huggingface.co/datasets/HarriziSaad/Nullomer
cd Nullomer

# Create a virtual environment (recommended)
python -m venv venv
source venv/bin/activate       # Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt
```

---

## Running the Analysis

### Option 1 — Full pipeline (recommended)

```bash
bash run_all.sh
```

This runs all six scripts in order and takes approximately **60–90 minutes** on standard hardware. The main bottleneck is the NEM scanning in script 02 (~30–60 min depending on CPU).

### Option 2 — Individual scripts

Run from the repo root. Scripts must be run in order as each depends on outputs from the previous one.

```bash
python scripts/01_nullomer_identification.py   # ~50 s
python scripts/02_nem_analysis.py              # ~30–60 min
python scripts/03_stress_element_analysis.py   # ~5 min
python scripts/04_thermodynamic_analysis.py    # ~2 min
python scripts/05_ml_and_network_analysis.py   # ~10 min
python scripts/06_statistical_synthesis.py     # ~1 min
```

---

## What Each Script Does

| Script | Analysis | Key output |
|---|---|---|
| `01` | Identifies all 11-mer nullomers in the *S. cerevisiae* R64-1-1 genome | `results/nullomers_k11.txt` |
| `02` | Finds NEMs across 26 ABC transporter genes and 1000 bp promoters | `results/nem_comprehensive_summary.csv` |
| `03` | Scans promoters for PDRE/STRE/HSE/AP-1; correlates with NEM density | `results/stress_element_nem_correlation.csv` |
| `04` | Computes ΔG and Tm for 10,000 nullomers vs random controls | `results/thermodynamic_summary.json` |
| `05` | Random Forest prediction, GP fitness landscape, STRING network | `results/ml_model_performance.json` |
| `06` | Hypothesis tests H1–H4 and Fisher's meta-analysis | `results/statistical_synthesis.json` |

---

## Expected Key Results

Verify your run matches the manuscript by checking these values in the output files:

### `results/nullomers_k11.txt`
- Line count: **463,220**

### `results/nem_comprehensive_summary.csv`
- Total NEM count (sum of `nem_count`): **174,799**
- Promoter mean NEM density: **1,331.9 NEMs/kb**
- Gene body mean NEM density: **1,085.7 NEMs/kb**
- Promoter enrichment over gene body: **22.6%**

### `results/stress_element_nem_correlation.csv`
- PDRE sites total: **30** across 8 genes
- STRE sites total: **228**
- HSE sites total: **2,540**
- AP-1 sites total: **36**

### `results/thermodynamic_summary.json`
```json
{
  "nullomer_Tm_mean": 41.73,
  "random_Tm_mean": 35.56,
  "nullomer_dG_mean": -13.96,
  "random_dG_mean": -12.13,
  "delta_dG_kcal_mol": 1.83,
  "boltzmann_fold_disadvantage": 19.4,
  "GC_Tm_pearson_r": 0.803,
  "pct_very_stable_nullomers": 99.7,
  "pct_hairpin": 22.4,
  "pct_g4": 1.0,
  "pct_imotif": 1.2
}
```

### `results/ml_model_performance.json`
```json
{
  "test_r2": 0.760,
  "test_rmse": 41.46,
  "cv_r2_mean": 0.717,
  "cv_r2_std": 0.045,
  "gp_r2": 0.896
}
```

### `results/statistical_synthesis.json`
```json
{
  "H2_PDRE_correlation": {
    "spearman_rho": 0.685,
    "spearman_p": 0.000111,
    "linear_slope_nems_per_pdre": 85.5
  },
  "H4_promoter_vs_gene": {
    "enrichment_pct": 22.6,
    "wilcoxon_p": 0.003
  },
  "meta_analysis_fishers": {
    "chi2_statistic": 51.32,
    "combined_p": 2.28e-08
  }
}
```

---

## Data Sources

All data is downloaded automatically at runtime. No manual downloads are needed.

| Resource | URL | Used by |
|---|---|---|
| *S. cerevisiae* genome R64-1-1 | Ensembl release 110 FASTA | scripts 01, 02, 03 |
| Gene annotations GFF3 | Ensembl release 110 GFF3 | scripts 02, 03 |
| Protein interactions | STRING v11.5 API (TaxID 4932, score ≥ 400) | script 05 |

---

## Troubleshooting

**Download failures** — If Ensembl is unreachable, retry or download manually:
```
https://ftp.ensembl.org/pub/release-110/fasta/saccharomyces_cerevisiae/dna/
https://ftp.ensembl.org/pub/release-110/gff3/saccharomyces_cerevisiae/
```
Place files in `data/` and rename to match the filenames used in the scripts.

**Memory errors during NEM scanning** — Script 02 holds the genome and nullomer set in memory simultaneously (~2–3 GB). Close other applications or run on a machine with more RAM.

**STRING API timeout** — Script 05 will catch the error and proceed with an empty network. The network results will be absent from `network_topology.csv` but all other outputs are unaffected.

**Missing dependencies** — Reinstall cleanly:
```bash
pip install --upgrade --force-reinstall -r requirements.txt
```

---

## Runtime Estimates

| Script | Approximate time |
|---|---|
| 01 Nullomer identification | 50 seconds |
| 02 NEM analysis | 30–60 minutes |
| 03 Stress element analysis | 5 minutes |
| 04 Thermodynamic analysis | 2 minutes |
| 05 ML and network analysis | 10 minutes |
| 06 Statistical synthesis | 1 minute |
| **Total** | **~60–90 minutes** |

Times measured on a standard laptop (8-core CPU, 16 GB RAM).

---

## Getting Help

If results do not match expected values:

1. Confirm you are running Python 3.8+ and have installed all packages from `requirements.txt`
2. Check that scripts are run from the **repo root**, not from inside `scripts/`
3. Open an issue at [https://huggingface.co/datasets/HarriziSaad/Nullomer/discussions](https://huggingface.co/datasets/HarriziSaad/Nullomer/discussions) with your Python version, OS, and the full error output

---

## Citation

```
Harrizi S., Nait Irahal I., Kabine M. (2025). Forbidden Words: How Nullomer
Constraints Channel Regulatory Evolution in ABC Transporters.
```