electricsheepafrica/africa-synth-population-ovarian-somatic-mutations-somalia

⚠️ Synthetic dataset — Parameterized from published SSA literature, not real observations. Not suitable for empirical analysis or policy inference.

Ovarian Cancer Somatic Mutations (Synthetic)

Dataset owner: Electric Sheep Africa
Dataset type: Synthetic ovarian cancer somatic mutations (drivers, CNV, MSI, HRD, platinum response)
Populations: Ovarian cancer patients from sub-Saharan Africa (modeled cohorts)
Version: 1.0.0
License: CC-BY-NC-4.0

---

1. Dataset Description

This dataset provides synthetic somatic mutation profiles for 30,000 ovarian tumors, with a focus on high-grade serous ovarian carcinoma (HGSOC) but including endometrioid, clear cell, mucinous, and other subtypes.

The dataset encodes:

• Per-sample driver mutation footprint (TP53, BRCA1/2, NF1, RB1, PTEN, PIK3CA, KRAS, ARID1A, CTNNB1, CDK12, ATM, ATR)
• Copy number variation (CNV) burden and key focal events (e.g., CCNE1, MYC, ERBB2 amplifications; PTEN, RB1, NF1 deletions)
• HRD-like score approximating homologous recombination deficiency patterns
• Microsatellite instability (MSI) status and score
• A simplified platinum_response label derived from HRD/BRCA and CNV features

All records are fully synthetic and parameterized using published ovarian cancer genomics studies (TCGA and others). No real patient-level data are used.

Important: This dataset contains no real patient data. It is derived entirely from literature-based distributions and internal coherence rules.

---

2. Intended Use

This dataset is intended for:

• Method development for somatic mutation–based risk models and response prediction
• Benchmarking algorithms for driver calling, HRD scoring, and CNV-aware modeling
• Educational use in ovarian cancer genomics and precision oncology

Not intended for clinical decision-making or individual patient stratification.

---

3. Cohort Design

3.1 Regions

• West_Africa
• East_Africa
• Southern_Africa
• Central_Africa

3.2 Sample Size & Demographics

• Total samples: 30,000 synthetic ovarian tumors
• Age range: 25–85 years, with most samples between 50–70
• Age groups: <35, 35-49, 50-64, 65+

3.3 Histologic Subtypes

• High_grade_serous (dominant subtype)
• Endometrioid
• Clear_cell
• Mucinous
• Other

The subtype mix is calibrated to approximate published distributions, with ~75% HGSOC.

3.4 Stage and Grade

• stage ∈ {I, II, III, IV} (enriched for advanced stage III/IV)
• grade ∈ {1, 2, 3} (enriched for high-grade tumors)

---

4. Somatic Mutations, CNV, and MSI

4.1 Driver Mutation Flags

Per-sample boolean flags indicate presence of at least one somatic variant in the gene:

• TP53_mut
• BRCA1_mut
• BRCA2_mut
• NF1_mut
• RB1_mut
• PTEN_mut
• PIK3CA_mut
• KRAS_mut
• ARID1A_mut
• CTNNB1_mut
• CDK12_mut
• ATM_mut
• ATR_mut

Mutation prevalences are tuned by histologic subtype (e.g., TP53 enriched in HGSOC; PIK3CA/ARID1A/KRAS enriched in endometrioid/clear cell/mucinous).

4.2 Tumor Mutational Burden (TMB)

• tmb_mut_per_mb – Tumor mutational burden (mutations per Mb) with modest elevation in MSI-H cases.

4.3 Copy Number Variation and HRD

Global CNV features:

• cnv_burden_fraction_genome_altered
• aneuploidy_score
• n_focal_amplifications
• n_focal_deletions
• hrd_score – HRD-like score (0–100)

Key focal events:

• CCNE1_amp, MYC_amp, ERBB2_amp
• PTEN_del, RB1_del, NF1_del

HGSOC samples are enriched for high CNV burden, high aneuploidy, and higher HRD scores relative to non-HGSOC subtypes.

4.4 Microsatellite Instability (MSI)

• msi_status ∈ {MSS, MSI-L, MSI-H}
• msi_score – continuous score (0–1) with higher values in MSI-L/MSI-H

MSI-H is rare (a few percent of tumors) but associated with higher TMB.

4.5 Platinum Response

• platinum_response ∈ {Sensitive, Resistant}

This label is simulated using a simple logistic model where BRCA1/2 mutations, high HRD score, and absence of CCNE1 amplification increase the probability of sensitivity.

---

5. Main File Schema

5.1 ovarian_somatic_data.csv

• Rows: 30,000 samples
• Columns (core variables):

Clinical & context:

• sample_id – Synthetic ID (OV_SAMPLE_00000 ...)
• region – {West_Africa, East_Africa, Southern_Africa, Central_Africa}
• age – Age in years
• age_group – {<35, 35-49, 50-64, 65+}
• histologic_subtype – {High_grade_serous, Endometrioid, Clear_cell, Mucinous, Other}
• stage – {I, II, III, IV}
• grade – {1, 2, 3}
• platinum_response – {Sensitive, Resistant}

Molecular features:

• msi_status, msi_score
• tmb_mut_per_mb
• Driver mutation flags listed above
• CNV burden and focal CNV flags listed above
• hrd_score

---

6. Data Access & Files

Main Dataset (root)

• ovarian_somatic_data.csv – 30,000 × ~35 variables (main table, CSV)
• ovarian_somatic_data.parquet – Same table in Parquet format for efficient loading.

Auxiliary Tables (local only)

The following tables are generated by the scripts but not distributed as separate files in this repository, to keep streaming simple:

• somatic_variants.csv – Long-format somatic variants per sample (gene-level details)
• cnv_segments.csv – Long-format CNV segments per sample

Key behaviors of these tables (driver footprints, CNV patterns) are summarized in the main table and the validation report.

---

7. Generation & Validation (Summary)

The dataset was generated using the same GENOMICS Synthetic Data Playbook principles as other projects in this repository:

1. Domain specification – Identify key ovarian cancer drivers, CNV patterns, MSI prevalence, and HRD behavior.
2. Literature-informed parameters – Approximate driver frequencies, CNV burden, HRD, and MSI from TCGA and other ovarian genomics studies (see local docs/LITERATURE_INVENTORY.csv).
3. Configuration – Encode parameters in ovarian_somatic/configs/ovarian_somatic_config.yaml.
4. Generation – Use scripts/generate_ovarian_somatic.py to simulate 30,000 tumors and auxiliary tables.
5. Validation – Use scripts/validate_ovarian_somatic.py to run structured checks on distributions, CNV/HRD coherence, MSI/TMB patterns, and missingness.

7.1 Validation Status

• Validation run on: ovarian_somatic/output/ovarian_somatic_data.csv
• Checks include:
  • Region and subtype distributions vs config
  • Driver mutation frequencies vs targets
  • CNV burden and HRD score vs histologic subtype
  • MSI prevalence vs config and TMB shift in MSI-H
  • Platinum response enrichment in HRD-high samples
  • Missingness in key variables

The current validation report indicates Overall Status: PASS.

---

8. Example Usage

8.1 Load with pandas

import pandas as pd

df = pd.read_csv("ovarian_somatic_data.csv")
print(df.shape)
print(df.head())

8.2 Load with datasets (Hugging Face)

from datasets import load_dataset

ov = load_dataset("electricsheepafrica/ovarian-somatic-mutations")
df = ov["train"].to_pandas()

print(f"Samples: {len(df)}")
print(f"Columns: {len(df.columns)}")

8.3 CNV burden vs histologic subtype

cnv = df.groupby("histologic_subtype")["cnv_burden_fraction_genome_altered"].describe()
print(cnv)

8.4 Platinum response vs HRD status

hrd_high = df["hrd_score"] > 42

sens_high = (df.loc[hrd_high, "platinum_response"] == "Sensitive").mean()
sens_low = (df.loc[~hrd_high, "platinum_response"] == "Sensitive").mean()

print("Platinum-sensitive fraction (HRD-high):", sens_high)
print("Platinum-sensitive fraction (HRD-low):", sens_low)

---

9. Ethical & Appropriate Use

9.1 Appropriate Uses ✅

• Method development and benchmarking in somatic mutation modeling
• Educational use in ovarian cancer genomics and HRD modeling
• Health equity and disparity analysis using synthetic data

9.2 Inappropriate Uses ❌

• Clinical decision-making for individual patients
• Real-world prognosis or treatment selection
• Insurance, employment, or financial decisions

This dataset is synthetic and must not be used as a substitute for real clinical or genomic data in patient care.

---

10. License

• License: CC-BY-NC-4.0
• Commercial use: Not permitted without explicit permission.

---

11. Citation

Electric Sheep Africa (2025).
Ovarian Cancer Somatic Mutations (Synthetic Dataset).
Generated using literature-informed synthetic data methods.
Hugging Face Datasets. Version 1.0.0.

---

12. Contact

• Organization: Electric Sheep Africa
• Hugging Face: https://huggingface.co/electricsheepafrica

Feedback and collaboration inquiries are welcome.