docs/DATA_DICTIONARY.md

# Data Dictionary

## Input Data

All input data is downloaded automatically by the scripts at runtime. Nothing needs to be uploaded manually.

### Genome & Annotation Files (cached to `data/` at runtime)

| File | Source | Size |
|---|---|---|
| `data/yeast_genome.fsa` | Ensembl R64-1-1 (release 110) | 12.16 Mb |
| `data/yeast.gff3.gz` | Ensembl release 110 | — |

Genome assembly: *S. cerevisiae* S288C, R64-1-1, 12,157,105 bp across 17 chromosomes including mitochondria.

---

## Output Data

All output files are written to `results/` by running the scripts in order.

### Nullomer List

**`results/nullomers_k11.txt`** — generated by `scripts/01_nullomer_identification.py`

- One 11-mer sequence per line, sorted alphabetically
- 463,220 sequences (11.04% of the 4,194,304 theoretical 11-mers)
- Mean GC content: 65.7% (vs 38.3% genome-wide)
- Both forward and reverse-complement strands are accounted for

---

### NEM Analysis

**`results/nem_comprehensive_summary.csv`** — generated by `scripts/02_nem_analysis.py`

| Column | Type | Description |
|---|---|---|
| `gene` | string | Gene name (e.g. PDR5) |
| `region` | string | `gene`, `promoter`, or `downstream` |
| `nem_count` | integer | Number of nullomer-emerging mutations |
| `seq_length` | integer | Length of the region in bp |
| `nem_density_per_kb` | float | NEMs per kilobase |
| `type` | string | Functional classification |
| `essential` | boolean | Gene essentiality |
| `stress` | boolean | Stress-responsive classification |
| `subfamily` | string | ABC transporter subfamily |

78 rows (26 genes × 3 regions). Total NEMs across all regions: 174,799.

**`results/nem_enrichment_analysis.csv`** — generated by `scripts/02_nem_analysis.py`

| Column | Type | Description |
|---|---|---|
| `gene` | string | Gene name |
| `region` | string | `gene`, `promoter`, or `downstream` |
| `observed_nems` | integer | Observed NEM count |
| `expected_nems` | float | Expected under Poisson null |
| `enrichment_ratio` | float | Observed / expected |
| `p_value` | float | Poisson p-value |
| `p_adjusted` | float | Bonferroni-corrected p-value |
| `significant` | boolean | p_adjusted < 0.05 |

**`results/stress_permutation_test.json`** — generated by `scripts/02_nem_analysis.py`

Permutation test results (10,000 iterations) comparing NEM density between stress-responsive and non-stress genes. Fields: `stress_mean`, `stress_std`, `nonstress_mean`, `nonstress_std`, `observed_diff_nems_per_kb`, `mannwhitney_u`, `mannwhitney_p`, `permutation_p`, `cohens_d`, `n_stress`, `n_nonstress`.

---

### Stress Element Analysis

**`results/stress_element_nem_correlation.csv`** — generated by `scripts/03_stress_element_analysis.py`

| Column | Type | Description |
|---|---|---|
| `gene` | string | Gene name |
| `promoter_length` | integer | Promoter length in bp (1000 bp) |
| `promoter_nems` | integer | NEM count in promoter |
| `nem_density_per_kb` | float | Promoter NEM density |
| `total_stress_elements` | integer | Sum of all binding sites |
| `PDRE` | integer | Pleiotropic Drug Response Element count |
| `STRE` | integer | Stress Response Element count |
| `HSE` | integer | Heat Shock Element count |
| `AP1` | integer | AP-1 element count |
| `type` | string | Functional classification |
| `essential` | boolean | Gene essentiality |
| `stress` | boolean | Stress-responsive classification |
| `is_drug_efflux` | boolean | Drug efflux gene flag |

26 rows (one per gene). Key result: PDRE count correlates with NEM density at Spearman ρ=0.685, p=1.1×10⁻⁴.

**`results/motif_disruption_by_nems.csv`** — generated by `scripts/03_stress_element_analysis.py`

| Column | Type | Description |
|---|---|---|
| `gene` | string | Gene name |
| `nem_position` | integer | Position in promoter sequence |
| `nem_mutation` | string | Mutation notation (e.g. A142G) |
| `element_type` | string | `PDRE`, `STRE`, `HSE`, or `AP1` |
| `motif_position` | integer | Motif start position |
| `motif_strand` | string | `+` or `-` |
| `position_in_motif` | integer | Position of NEM within the motif |

Records where a single mutation both creates a nullomer and falls inside a known TF binding site. Total: 16,480 disruptions across all elements.

---

### Thermodynamic Analysis

**`results/nullomer_thermodynamics.csv`** — generated by `scripts/04_thermodynamic_analysis.py`

| Column | Type | Description |
|---|---|---|
| `sequence` | string | 11-mer sequence |
| `group` | string | `nullomer` or `random` |
| `Tm` | float | Melting temperature (°C) |
| `dG` | float | Gibbs free energy at 37°C (kcal/mol) |
| `GC` | float | GC fraction (0–1) |
| `hairpin` | boolean | Palindromic hairpin potential |
| `g4` | boolean | G-quadruplex motif (GGGG) present |
| `imotif` | boolean | i-motif motif (CCCC) present |

20,000 rows (10,000 nullomers + 10,000 random controls). Parameters: SantaLucia (1998) nearest-neighbour, 37°C, 1 M NaCl.

**`results/thermodynamic_summary.json`** — generated by `scripts/04_thermodynamic_analysis.py`

Key values confirmed against the manuscript:

| Metric | Nullomers | Random |
|---|---|---|
| Mean Tm | 41.73 ± 5.70 °C | 35.56 ± 6.63 °C |
| Mean ΔG | −13.96 ± 1.52 kcal/mol | −12.13 ± 1.80 kcal/mol |
| ΔΔG | 1.83 kcal/mol | — |
| Boltzmann fold disadvantage | 19.4× | — |
| GC–Tm Pearson r | 0.803 | — |
| Very stable (ΔG < −10) | 99.7% | — |
| Hairpin potential | 22.4% | — |
| G-quadruplex | 1.0% | — |
| i-motif | 1.2% | — |

---

### ML and Network Analysis

**`results/ml_feature_importance.csv`** — generated by `scripts/05_ml_and_network_analysis.py`

| Column | Type | Description |
|---|---|---|
| `feature` | string | Feature name |
| `importance` | float | Random Forest mean decrease in impurity |

Top features: `at_content` (0.359), `gc_content` (0.356), `cg_dinuc` (0.153).

**`results/ml_model_performance.json`** — generated by `scripts/05_ml_and_network_analysis.py`

Random Forest performance (100 bp windows, 50 bp step, 26 genes):

| Metric | Value |
|---|---|
| Test R² | 0.760 |
| Test RMSE | 41.46 NEMs |
| CV R² (5-fold) | 0.717 ± 0.045 |

Also contains Gaussian Process fitness landscape results: R²=0.896, RMSE=93.8 NEMs/kb.

**`results/network_topology.csv`** — generated by `scripts/05_ml_and_network_analysis.py`

| Column | Type | Description |
|---|---|---|
| `gene` | string | Gene name |
| `nem_density` | float | NEM density (NEMs/kb) |
| `degree` | integer | Number of STRING interaction partners |
| `betweenness` | float | Betweenness centrality |
| `closeness` | float | Closeness centrality |
| `eigenvector` | float | Eigenvector centrality |
| `is_drug_efflux` | boolean | Drug efflux gene flag |

Network: 26 nodes, 13 edges (STRING v11.5, score ≥ 400, physical interactions only).

**`results/fragility_scores.csv`** — generated by `scripts/05_ml_and_network_analysis.py`

| Column | Type | Description |
|---|---|---|
| `gene` | string | Gene name |
| `fragility_score` | float | F = 0.4×(NEM/5000) + 0.3×(degree/n) + 0.3×(neighbor_NEM/5000) |

Top 5: PDR15 (1.402), PDR10 (1.330), PDR5 (1.238), SNQ2 (1.161), PDR12 (1.076).

---

### Statistical Synthesis

**`results/statistical_synthesis.json`** — generated by `scripts/06_statistical_synthesis.py`

Contains all four hypothesis tests and Fisher's combined p-value:

| Test | Result |
|---|---|
| H1: Stress vs non-stress NEM density | Mann-Whitney p=0.019, permutation p=0.006, d=1.36 |
| H2: PDRE–NEM correlation | Spearman ρ=0.685, p=1.1×10⁻⁴, slope=85.5 NEMs/kb per PDRE |
| H3: Drug efflux vs other | Mann-Whitney p=0.018, Cohen's d=1.08 |
| H4: Promoter vs gene body density | Wilcoxon p=0.003, enrichment=22.6% |
| Meta-analysis (Fisher) | χ²=51.32, df=8, combined p=2.28×10⁻⁸ |

---

## ABC Transporters Analyzed

26 genes spanning: drug efflux pumps (PDR5, SNQ2, YOR1, PDR10, PDR11, PDR12, PDR15, PDR18, YCF1), transcriptional regulators of drug resistance (PDR1, PDR3, PDR16, PDR17), mitochondrial transporters (ATM1, MDL1, MDL2), translation-related (YEF3, GCN20, ARB1, RLI1), and others (VMR1, YBT1, BPT1, HMT1, NMD5, STE6).

Promoter length used throughout: **1000 bp upstream** of each start codon.

---

## File Formats

All CSV files use comma separation, UTF-8 encoding, and a header row. JSON files use UTF-8 with two-space indentation. The nullomer list (`nullomers_k11.txt`) has one 11-mer per line, sorted lexicographically.