electricsheepafrica/ssa-breast-rna-expression

SSA Breast RNA Expression Dataset (Normal & Tumor, Multi-ancestry, Synthetic)

Dataset summary

This dataset provides a synthetic RNA expression cohort spanning normal breast tissue and invasive breast cancer across multiple ancestry groups, with emphasis on Sub-Saharan Africa (SSA) and reference populations.

Each sample has:

• Demographics (population, sex, age, SSA indicator).
• Tissue context: Normal_breast vs Tumor.
• Tumor PAM50-like subtype for tumour samples (Luminal_A, Luminal_B, HER2_enriched, Basal_like, Normal_like).
• Summary RNA-seq metrics (library size, number of detected genes).
• Grouped gene expression (log2(TPM+1) scale) for:
  • Housekeeping genes (ACTB, GAPDH, RPLP0).
  • ER signalling (ESR1, PGR, GATA3, FOXA1).
  • Proliferation markers (MKI67, CCNB1, BIRC5).
  • HER2 pathway (ERBB2).
  • Basal markers (KRT5, KRT14, KRT17, EGFR).
  • Immune/ancestry-associated genes (IL20RA, MARCO, HLA_DPA1, HLA_DPB1).
  • Ancestry-associated protein-coding genes (SORD, FAM3A, TAS2R43, NUTM2F, PWP2, PM20D1, BIRC7).
  • Oncogenic lncRNAs (HOTAIR, MALAT1, NEAT1, AC009283_1) and XIST.
  • OncomiR / tumour-suppressor microRNAs (MIR21, MIR155, MIR10B, MIR200C, MIR34A).

Expression values are simulated from log2(TPM+1)-like distributions calibrated using GTEx normal breast, TCGA-BRCA tumour/normal, ancestry-associated RNA-seq studies (e.g. Ping et al. 2020, SCCS & TCGA), and core miRNA/lncRNA breast cancer literature. All records are fully synthetic and non-identifiable.

Cohort design

Sample size and populations

• Total N: 10,000 synthetic breast samples.

• Populations:

  • SSA_West: 2,000
  • SSA_East: 2,000
  • SSA_Central: 1,500
  • SSA_Southern: 1,500
  • AAW (African American women): 1,500
  • EUR (European reference): 1,000
  • EAS (East Asian reference): 500

• Sex:

  • Predominantly Female, with ~1% male breast samples.

• Age:

  • 18–90 years.
  • Slightly younger mean age in SSA cohorts vs older in AAW/EUR/EAS, consistent with cancer registry patterns.

Tissue and subtype design

• Tissue type (tissue_type):

  • Normal_breast
  • Tumor

  Normal breast profiles are conceptually anchored to GTEx normal breast RNA-seq distributions (TPM-based), while tumour profiles follow TCGA-BRCA-like patterns.

• Tumor subtypes (tumor_subtype, for Tumor only):

  • Luminal_A
  • Luminal_B
  • HER2_enriched
  • Basal_like
  • Normal_like

  Subtype distributions by population mirror published PAM50 patterns: higher Basal_like fractions in SSA and African American women, and more Luminal_A disease in EUR/EAS reference cohorts.

Expression features

Summary RNA-seq metrics

Variables:

• library_size_millions – approximate library size in millions of reads.
• detected_genes – number of genes above a minimal expression threshold.

Anchoring:

• GTEx and TCGA RNA-seq libraries typically range 30–80M reads; the synthetic dataset uses means around 50M (normal) and 60M (tumour) with realistic variance.
• Tumour samples have slightly more detected genes on average than normal breast, reflecting increased transcriptional diversity.

Housekeeping and global scaling

• ACTB, GAPDH, RPLP0 are modeled as stable housekeeping genes with high expression and low variance across tissues and populations.
• Values are generated around log2(TPM+1) ≈ 10–11 with small SD, providing internal calibration points.

ER signalling genes

Variables:

• ESR1, PGR, GATA3, FOXA1

Anchoring:

• TCGA and panel studies show high ER-pathway expression in Luminal_A/B, intermediate in HER2_enriched, and low in Basal_like tumours.
• Normal breast tissue has intermediate ESR1/PGR with heterogeneity across individuals.

In the synthetic dataset:

• Luminal_A tumours have the highest ER-pathway expression.
• Basal_like tumours show markedly reduced ESR1/PGR.
• Normal breast is modeled with ER expression between Luminal_A tumours and Basal_like tumours.

Proliferation markers

Variables:

• MKI67, CCNB1, BIRC5

Anchoring:

• Proliferation signatures (e.g. Ki-67, cell-cycle genes) are markedly higher in Basal_like and HER2_enriched tumours, intermediate in Luminal_B, and lowest in Luminal_A and normal breast.

The dataset encodes higher log2(TPM+1) means for these markers in aggressive subtypes, especially Basal_like.

HER2 pathway

Variable:

• ERBB2

Anchoring:

• HER2_enriched tumours have strongly upregulated ERBB2 compared with other subtypes and normal tissue.

The synthetic distribution reproduces this pattern with large subtype-specific mean shifts.

Basal epithelial markers

Variables:

• KRT5, KRT14, KRT17, EGFR

Anchoring:

• Basal-like and many TNBC tumours express high basal cytokeratins and EGFR, whereas luminal tumours and normal epithelium show lower levels.

In the dataset:

• Basal_like tumours have the highest basal-marker expression.
• Luminal subtypes show much lower values.
• Normal breast sits in an intermediate/low range.

Immune / ancestry-associated genes

Variables:

• IL20RA, MARCO, HLA_DPA1, HLA_DPB1

Anchoring:

• Studies such as Ping et al. (Carcinogenesis 2020) and multi-ancestry RNA-seq analyses have identified immune and antigen-presentation genes differing between African- and European-ancestry tumours.

In the synthetic data:

• These genes have higher mean expression in tumours from African ancestry cohorts (SSA_*, AAW), particularly in Basal_like and HER2_enriched subtypes.

Ancestry-associated protein-coding genes

Variables:

• African-ancestry–high: SORD, FAM3A, TAS2R43, NUTM2F, PWP2.
• European/East-Asian–high: PM20D1, BIRC7.

Anchoring:

• Ping et al. identified a 59-gene signature that discriminates African American vs European American tumours, including genes such as TAS2R43, SORD, FAM3A, and BIRC7.

The dataset encodes:

• Higher mean expression of African-ancestry–high genes in SSA_* and AAW populations vs EUR/EAS.
• Higher PM20D1/BIRC7 in EUR/EAS, with modestly lower levels in African ancestry groups.

Long non-coding RNAs (lncRNAs)

Variables:

• Oncogenic lncRNAs: HOTAIR, MALAT1, NEAT1, AC009283_1.
• X-linked dosage lncRNA: XIST.

Anchoring:

• Multiple lncRNA profiling studies show upregulation of HOTAIR, MALAT1, NEAT1, AC009283.1 in breast tumours vs normal tissue, with subtype-specific patterns.
• XIST expression reflects X-chromosome inactivation and is modeled as relatively stable across tissues.

In the synthetic data:

• Oncogenic lncRNAs are substantially higher in tumours (especially Basal_like and HER2_enriched) vs normal breast.
• Validation enforces that most of these lncRNAs have higher mean expression in tumour than normal.

MicroRNAs

Variables (modeled on a log-transformed abundance scale):

• OncomiRs: MIR21, MIR155, MIR10B.
• Tumour suppressor miRNAs: MIR200C, MIR34A.

Anchoring:

• Classical breast cancer miRNA studies (e.g. Iorio et al., Cancer Res 2005) report overexpression of miR-21, miR-155, miR-10b and downregulation of miR-34a and miR-200 family in tumours vs normal breast.

The synthetic dataset encodes:

• Higher oncomiR expression in tumour vs normal, especially in Basal_like and HER2_enriched.
• Lower MIR200C/MIR34A in tumours vs normal tissue.

File and schema

rna_expression_data.parquet / rna_expression_data.csv

Each row represents a synthetic RNA-seq sample with demographics, tissue context, and grouped expression values.

Demographics and context

• sample_id
• population (SSA_West, SSA_East, SSA_Central, SSA_Southern, AAW, EUR, EAS)
• region (derived from population)
• is_SSA (boolean)
• is_reference_panel (boolean; AAW, EUR, EAS)
• sex
• age
• tissue_type (Normal_breast, Tumor)
• tumor_subtype (PAM50-like; Normal_breast samples are labelled Normal_like)
• is_tumor (boolean)
• library_size_millions
• detected_genes

Expression features (log2(TPM+1)-like)

• Housekeeping: ACTB, GAPDH, RPLP0
• ER signalling: ESR1, PGR, GATA3, FOXA1
• Proliferation: MKI67, CCNB1, BIRC5
• HER2 pathway: ERBB2
• Basal markers: KRT5, KRT14, KRT17, EGFR
• Immune/ancestry genes: IL20RA, MARCO, HLA_DPA1, HLA_DPB1
• Ancestry-associated genes: SORD, FAM3A, TAS2R43, NUTM2F, PWP2, PM20D1, BIRC7
• lncRNAs: HOTAIR, MALAT1, NEAT1, AC009283_1, XIST
• microRNAs (log-like abundance): MIR21, MIR155, MIR10B, MIR200C, MIR34A

Generation

The dataset is generated using:

• rna_expression/scripts/generate_rna_expression.py

with configuration in:

• rna_expression/configs/rna_expression_config.yaml

and literature inventory in:

• rna_expression/docs/LITERATURE_INVENTORY.csv

Key generation steps:

1. Cohort sampling – multi-ancestry sampling of populations, sex, and age; assignment of Normal_breast vs Tumor by population-specific fractions.
2. Subtype assignment – for tumour samples, subtype assignment using population-specific PAM50-like distributions (higher Basal_like in SSA/AAW).
3. Summary RNA-seq metrics – simulation of library_size_millions and detected_genes by tissue.
4. Grouped gene expression – for each gene group, sample log2(TPM+1) using:
   • Base mean/SD.
   • Additive effects of tissue type (normal vs tumour).
   • Additive effects of tumour subtype (e.g., higher MKI67 in Basal_like).
   • Additive ancestry effects for specified gene sets (e.g., higher SORD in SSA_* and AAW).

Validation

Validation is performed with:

• rna_expression/scripts/validate_rna_expression.py

and summarized in:

• rna_expression/output/validation_report.md

Checks include:

• C01–C02 – Sample size and population counts vs configuration.
• C03 – Tissue-type distributions by population.
• C04 – Tumour subtype distributions among tumour samples, by population.
• C05 – Ancestry-associated gene signature differences (African/SSA vs EUR/EAS) for configured genes.
• C06 – Oncogenic lncRNAs higher in tumour vs normal breast.
• C07 – OncomiR microRNAs higher in tumour vs normal breast.
• C08 – Missingness across key columns (demographics, context, summary metrics, expression features).

The released configuration targets a 10% absolute deviation tolerance for categorical distributions and ensures that configured lncRNA/miRNA and ancestry-signature patterns hold for the majority of genes.

Intended use

This synthetic dataset is intended for:

• Method development for RNA-seq modeling, subtype classification, and ancestry-aware modeling.
• Teaching/education on:
  • Differences between normal breast and tumour expression profiles.
  • Relationships between ER status, proliferation, HER2, basal signatures, and ancestry.
  • Integration of lncRNA and miRNA information into gene expression models.
• Benchmarking multi-omics pipelines when combined with other Electric Sheep Africa modules (e.g., methylation, immune profiles, pathology).

It is not intended for:

• Estimating real-world gene expression distributions or subtype frequencies in any specific population.
• Direct clinical decision-making, biomarker cut-off selection, or patient-level inference.

Ethical considerations

• All data here are synthetically generated; no real patient-level expression, genotype, or clinical records are used.
• Ancestry-associated patterns are derived from published group-level findings and should not be used to essentialize or stereotype populations.
• Users should interpret modeled disparities alongside rigorous epidemiological and socio-environmental data.

License

• License: CC BY-NC 4.0.
• Free for non-commercial research, method development, and education with attribution.

Citation

If you use this dataset, please cite:

Electric Sheep Africa. "SSA Breast RNA Expression Dataset (Normal & Tumor, Multi-ancestry, Synthetic)." Hugging Face Datasets.

and, where appropriate, foundational resources such as GTEx, TCGA-BRCA, and ancestry-focused RNA-seq and miRNA/lncRNA studies in breast cancer.