electricsheepafrica/ssa-snp-array-aims

Genome-wide SNP Array with Ancestry Informative Markers (SSA-focused)

Abstract

This synthetic dataset provides genome-wide single nucleotide polymorphism (SNP) genotypes for 5,000 women designed to approximate the structure of a commercial 500K–1M SNP array. The cohort is centered on sub-Saharan African (SSA) women sampled across major regions, with additional reference panels of African American women (AAW) and women of European and East Asian ancestry. The SNP panel combines a genome-wide backbone with an enriched set of ancestry informative markers (AIMs) capable of distinguishing SSA populations from other continental groups and capturing African American admixture.

The dataset is intended for benchmarking and methods development in ancestry inference, population structure analysis, quality control pipelines, and admixture-aware association methods, while avoiding any use of identifiable individual-level real genotypes. All parameters are encoded in a YAML configuration file and linked to a structured LITERATURE_INVENTORY.csv, following the GENOMICS Synthetic Data Playbook.

Background and Rationale

Genomic studies in sub-Saharan Africa remain underrepresented despite the continent harboring the greatest human genetic diversity. Standard genome-wide SNP arrays and reference panels have historically been optimized for European-ancestry populations, with limited coverage of African-specific variation and population structure. This limits fine-scale ancestry inference, imputation accuracy, and transferability of association findings to African and African-descended populations.

Ancestry informative markers (AIMs) are SNPs selected to show large allele-frequency differences between populations (high informativeness or Fst). Carefully designed AIM panels, combined with dense genome-wide SNP backbones, enable:

• Continental ancestry inference (e.g., SSA vs European vs East Asian).
• Detection of population outliers and stratification in case–control studies.
• Modeling of admixed populations, such as African Americans.
• Educational and methods-focused demonstrations of principal component analysis (PCA), clustering, and admixture estimation.

This dataset provides a synthetic yet literature-anchored approximation of such an array, emphasizing SSA regional diversity and comparison to AAW, European, and East Asian panels.

Cohort Design

Populations and Sample Sizes

The dataset comprises 5,000 simulated female individuals partitioned as follows:

• SSA women (core cohort, 3,000)
  • West Africa (SSA_West): 1,000
  • East Africa (SSA_East): 800
  • Central Africa (SSA_Central): 600
  • Southern Africa (SSA_Southern): 600
• Reference panels (2,000)
  • African American women (AAW): 1,000
  • European women (EUR): 500
  • East Asian women (EAS): 500

All individuals are encoded as Female in the configuration to match the design focus on women.

Ancestry Modeling

• SSA regional groups are modeled as primarily African-ancestry populations with region-specific labels (West, East, Central, Southern). Allele frequencies are drawn from an "AFR-like" spectrum representing common patterns seen in African populations on high-density SNP arrays.
• European (EUR) and East Asian (EAS) panels are modeled using distinct allele-frequency spectra to create strong between-continental differentiation suitable for PCA and Fst analyses.
• African American women (AAW) are modeled as an admixed group, with individual-level ancestry proportions drawn from a distribution centered on approximately 80% African and 20% European ancestry, with variation across individuals. This is consistent with published genome-wide admixture analyses of African Americans using 500K SNP arrays and AIM panels.

SNP Panel Design

Genome-wide Backbone

• Target size: approximately 550,000 SNPs.
• Purpose: represent a generic high-density autosomal SNP backbone similar in scale to commercial arrays (500K–1M SNPs).
• For each simulated SNP, ancestry-specific allele frequencies are drawn from Beta distributions that approximate realistic minor-allele-frequency (MAF) spectra in African, European, and East Asian reference populations.
• The backbone is not tied to specific physical coordinates; instead, it represents a conceptual genome-wide panel suitable for methods benchmarking.

Ancestry Informative Markers (AIMs)

• Additional 50,000 SNPs are designated as ancestry informative markers.
• AIMs are simulated to have larger allele-frequency contrasts between continental ancestries (e.g., SSA vs EUR vs EAS), informed by published AIM panels such as the 93-SNP set validated by Kosoy et al. (BMC Genet. 2009;10:39).
• A subset of AIMs is configured to provide substructure information within SSA, enabling separation of some regional labels (e.g., Pygmy vs other SSA populations in the motivating literature).
• The configuration also specifies a smaller AIMs-only subset (e.g., 5,000 SNPs) to support lightweight teaching examples and rapid PCA/UMAP visualizations.

Total SNP Count

• The configuration targets a total of 600,000 SNPs:
  • ~550,000 backbone SNPs.
  • ~50,000 AIM SNPs.
• The exact implemented number is defined by the YAML configuration and may be modestly adjusted in future versions while maintaining the conceptual 500K–1M array scale.

Simulation Framework

The synthetic genotypes are generated in a two-stage process, driven entirely by a YAML configuration file (snp_array_config.yaml) and literature-backed hyperparameters.

1. Allele-frequency generation

   • For each SNP and each ancestral population model (African-like, European-like, East Asian-like), a reference allele frequency is sampled from a Beta distribution.
   • SSA regional groups (SSA_West, SSA_East, SSA_Central, SSA_Southern) are assigned the AFR-like frequency model.
   • EUR and EAS panels are assigned EUR-like and EAS-like models, respectively.
   • AAW group frequencies are derived from an admixture model combining SSA_West and EUR contributions.

2. Genotype generation

   • For each individual, diploid genotypes are drawn under Hardy–Weinberg equilibrium given the population-specific allele frequencies.
   • For AAW individuals, the effective allele frequency at each SNP is a weighted combination of African and European reference frequencies based on the individual's simulated ancestry proportion.
   • AIM SNPs are simulated to have higher between-group differentiation (higher Fst) than backbone SNPs by drawing from more extreme Beta distributions and/or by explicitly shifting allele frequencies between ancestries.

No real genotypes are used at any point. All draws originate from parametric distributions or discrete mixture models informed by published summaries, not by individual-level datasets.

File Structure and Data Records

The project is organized similarly to other datasets in this repository:

• snp_array/configs/snp_array_config.yaml — complete configuration of sample sizes, population labels, allele-frequency models, AIM parameters, admixture model, and output paths.
• snp_array/docs/LITERATURE_INVENTORY.csv — literature inventory linking key parameters (AIM panel properties, population structure assumptions) to specific citations.
• snp_array/output/ — generated synthetic data files.
• snp_array/huggingface_dataset/ — files prepared for distribution on the Hugging Face Hub (README, Parquet/CSV artifacts, dataset script if used).

Sample metadata

Planned main sample-level table (Parquet with a small CSV mirror):

• sample_id — unique identifier (e.g., SNP_SAMPLE_000001).
• population — categorical label (SSA_West, SSA_East, SSA_Central, SSA_Southern, AAW, EUR, EAS).
• region — SSA region label for SSA samples; Non_SSA otherwise.
• is_SSA — boolean indicator for SSA samples.
• is_reference_panel — boolean indicator for reference-panel individuals (AAW, EUR, EAS).
• sex — fixed to Female for all individuals.
• eur_ancestry_prop — simulated proportion of European ancestry (relevant for AAW; ~0 for most SSA samples; ~1 for EUR samples).

Variant metadata

Planned variant-level table (Parquet):

• variant_id — synthetic identifier for the SNP (e.g., var_000001).
• panel — backbone or AIM.
• is_aim — boolean flag for AIM SNPs.
• maf_SSA_overall — minor allele frequency across all SSA samples.
• maf_EUR — minor allele frequency in European panel.
• maf_EAS — minor allele frequency in East Asian panel.
• Optional: per-region SSA MAF fields (e.g., maf_SSA_West, maf_SSA_East), depending on final validation requirements.

Genotype matrices

To keep the dataset scalable and streaming-friendly, genotypes are stored in columnar formats:

• snp_array_genotypes_chr*.parquet — variant-major genotype matrices, typically chunked by chromosome or blocks, with each row representing a SNP and each column (or nested field) representing genotypes across samples.
• snp_array_genotypes_aims.parquet — compact AIMs-only subset to facilitate rapid analyses.

Exact schemas are defined in the generator script and mirrored in the Hugging Face dataset card once the first reference generation is completed.

Technical Validation

A dedicated validation script will be provided to evaluate the following aspects of each generation run:

• Population sample sizes — check that realized counts per population match configuration targets within a small tolerance.
• MAF spectra by population — compare empirical distributions of minor allele frequencies in SSA, AAW, EUR, and EAS panels to the shapes implied by the Beta hyperparameters.
• AIM information content — verify that AIM SNPs show consistently higher between-group differentiation (e.g., via Fst or related statistics) than backbone SNPs.
• Continental separation and SSA vs reference panels — use PCA or similar methods to confirm that SSA vs EUR vs EAS panels and the AAW group separate in expected ways (AAW typically intermediate between SSA_West and EUR).
• Admixture distribution in AAW — confirm that the empirical distribution of eur_ancestry_prop in the AAW cohort matches the configuration’s target mean and variance.
• Missingness checks — verify that core variables (population labels, sex, ancestry proportions) have negligible missingness.

Validation results will be summarized in a Markdown report (e.g., snp_array/output/validation_report.md) using the same PASS/WARN/FAIL conventions as other datasets in this repository.

Intended Use

This synthetic SNP dataset is designed for:

• Methods development and benchmarking in:
  • Ancestry inference and supervised classification of continental origin.
  • Admixture estimation and visualization (e.g., STRUCTURE/ADMIXTURE-like methods).
  • Population stratification correction in GWAS pipelines.
  • PCA/UMAP-based quality control and outlier detection.
• Teaching and training in population genetics, focusing on African and African-descended populations.
• Prototyping of pipelines that will eventually be run on real SSA genotype data, without exposing any individual-level real genotypes during development.

This dataset is not intended for:

• Clinical decision making.
• Direct inference of individual ancestry for any real-world person.
• Derivation of health-related risk scores for individuals or groups.

Limitations

• The dataset is fully synthetic and does not attempt to reproduce the exact linkage disequilibrium (LD) structure, haplotype blocks, or rare variant spectrum of any specific real array.
• Allele-frequency and admixture parameters are anchored in summaries from the literature, not in direct reuse of real genotypes.
• Fine-scale substructure within SSA regions is simplified into a small number of labeled groups and may not capture all known ethno-linguistic or geographic patterns.
• Functional annotation of SNPs (e.g., coding vs non-coding, regulatory annotations) is not modeled; all SNPs are treatment-neutral markers.

Ethical Considerations

Although this dataset contains no real genotypes, ancestry and population labels remain sensitive. Users should:

• Avoid stigmatizing language or interpretations when discussing differences between groups.
• Use the dataset to improve inclusivity and robustness of genomic methods, particularly for SSA populations and African-descended groups.
• Be cautious when extrapolating algorithm performance from synthetic data to real-world cohorts, especially in clinical or policy contexts.

Provenance and Citation

The SNP array configuration and AIM panel design draw on published literature, including but not limited to:

1. Kosoy R et al. An ancestry informative marker set for determining continental origin: validation and extension using human genome diversity panels. BMC Genet. 2009;10:39.
2. Genome-wide SNP and admixture studies of West Africans and African Americans using 500K arrays and AIM panels (e.g., Bryc et al., PNAS 2010), informing approximate African/European ancestry proportions in African American populations.

The snp_array/docs/LITERATURE_INVENTORY.csv file provides a structured mapping between specific parameters in the configuration and the literature used to motivate them.

If you use this dataset in academic work, please cite the hosting repository (Electric Sheep Africa) and this dataset by name, and, where appropriate, also cite the underlying methodological and population-genetics references listed above.