Results for ADAMIXTURE: Ancestry Proportions (Q) and Allele Frequencies (P) on Simulated and Public Empirical Datasets

Overview: This repository contains the full inference results supporting the study ADAMIXTURE: First-Order Optimization for Biobank-Scale Ancestry Inference. It includes the estimated admixture proportions (Q matrices) and ancestral allele frequencies (P matrices) generated by ADAMIXTURE and four competing state-of-the-art methods. Methods Benchmarked: The outputs are provided for the following algorithms, all executed using default hyperparameters as described in the manuscript:ADAMIXTURE (Proposed method)ADMIXTURE (v1.4.1)fastmixture (v1.2.0)Neural Admixture (v1.4.4)SCOPEDatasets included: The repository covers results for both synthetic and public empirical datasets: Simulated Scenarios (A–D): Standardized benchmarks modeling discrete populations (A-C) and complex admixture events (D).Human Genome Diversity Project (HGDP): A global reference dataset comprising ~1,000 individuals from 51 populations.Domestic Dog Dataset (Dogs): Genotype data from 1,355 individuals spanning 166 breeds, representing strong genetic drift.File Structure Format: For each dataset and method, two files are provided: *.Q: The N x K matrix of ancestry proportions, where row i corresponds to individual i.*.P: The M x K matrix of ancestral allele frequencies, where row j corresponds to SNP j.The model complexity (K) used for each dataset is as follows:Dataset A: K=3Dataset B: K=4Dataset C: K=5Dataset D: K=3HGDP: K=8Dogs: K=20Data availability note: Results for the UK Biobank (UKB) dataset are excluded from this public repository due to data privacy and material transfer agreement (MTA) restrictions. Researchers approved to access UK Biobank data may reproduce those results using the code available at: https://github.com/ai-sandbox/adamixture.