Efficient analysis of large datasets and sex bias with ADMIXTURE

Suyash Shringarpure, Carlos D. Bustamante, Kenneth Lange, David H. AlexanderbioRxiv 2016 · DOI

TL;DR

This paper introduces two significant enhancements to the ADMIXTURE software, making it faster and more versatile for analyzing large genomic datasets. The first allows for efficient ancestry inference of new individuals by "projecting" them onto a pre-existing population structure, achieving a 5x speedup. The second enables ADMIXTURE to correctly analyze haploid sex chromosomes, improving the detection of sex-biased admixture.

What Problem Does This Paper Address?

The field of genomics is rapidly generating massive datasets, presenting a computational challenge for existing analysis tools. ADMIXTURE is a popular program for inferring individual ancestry from genetic data, but its efficiency on very large datasets and its ability to handle sex-specific genetic information (like the X chromosome) were limitations. Specifically, analyzing new individuals required re-running ADMIXTURE on the entire dataset, which is time-consuming. Furthermore, sex-biased admixture, where different ancestral contributions come from males and females, is a crucial aspect of population genetics, but accurately detecting it was hindered by ADMIXTURE's traditional focus on diploid autosomal data.

Methods

The authors introduce two key extensions to ADMIXTURE:

  • Projection Mode: This new mode allows ADMIXTURE to infer ancestry for a new set of individuals by using pre-computed ancestral population allele frequencies (learned from a reference panel or a previous ADMIXTURE run). Instead of re-calculating everything, it efficiently projects the new individuals onto the existing population structure. This is analogous to how principal component analysis (PCA) can be used to project new data onto existing principal components.
  • Haploid Sex Chromosome Handling: ADMIXTURE is modified to correctly model the log-likelihood for haploid chromosomes, specifically the X chromosome in males. This involves adjusting the calculations to account for the fact that males have only one copy of the X chromosome, unlike the two copies found in females and on autosomes. This is implemented via a --haploid option that specifies the sex and chromosome.

Datasets

  • 1000 Genomes Project: This dataset was extensively used for both training and evaluation. Specifically, data from Phase 1 was utilized to create a large dataset of 10,920 individuals for testing the projection method and for analyzing sex bias in African-Americans.
  • HapMap Project: Used in conjunction with the 1000 Genomes Project to demonstrate the utility of the new extensions.
  • Exome Aggregation Consortium (ExAC) data: Used as a reference for evaluating the accuracy of allele frequency estimates.

Key Insights and Ideas

  • Computational Efficiency is Key: The "projection" method dramatically speeds up ancestry inference for large datasets, making it feasible to analyze datasets that were previously computationally prohibitive.
  • Leveraging Existing Structure: By allowing users to project new samples onto established population structures, ADMIXTURE becomes more flexible and can be integrated into existing analytical pipelines.
  • Sex Matters in Ancestry: The ability to analyze sex chromosomes opens up new avenues for understanding population history, particularly the phenomenon of sex-biased admixture, which can reveal different demographic processes for males and females.
  • Improved Sex Bias Detection: Analyzing the X chromosome in males, in addition to autosomes, significantly increases the power to detect sex-biased admixture in populations.

Key Results

  • 5x Speedup: The projection method achieved a 5x speedup in running time when inferring ancestry for 10,920 individuals compared to running ADMIXTURE unsupervised on the entire dataset. This speedup increases with the number of ancestral clusters (K).
  • Identical Ancestry Proportions: The ancestry proportions inferred using the projection method were found to be identical to those obtained from the unsupervised approach.
  • Faster than iAdmix: The ADMIXTURE projection implementation was found to be approximately 4 times faster than the independently developed iAdmix projection method.
  • Improved Sex Bias Detection: The analysis of African-Americans demonstrated that including male samples and analyzing the X chromosome with the --haploid option allowed for the rejection of the null hypothesis of no sex bias, a result not achieved when analyzing females only.

Limitations and Future Work

  • Novel Populations in Projection: A limitation of the projection approach is that if the projected data contains a novel population not present in the initial training set, the results may not be identical to running ADMIXTURE on the combined dataset.
  • Related Individuals and Small Pedigrees: For related individuals, if the number of founders in a pedigree is small, the error in allele frequency estimates from the unrelated individuals might be large, potentially limiting the accuracy of the projection approach in such cases.
  • Future Work: The authors suggest that these extensions will make ADMIXTURE more useful for researchers as genomic sequencing and individual-level genotype data continue to grow. The improved efficiency and versatility are expected to facilitate deeper insights into population genetics and human history.

Summary updated

Abstract

Abstract Background : A number of large genomic datasets are being generated for studies of human ancestry and diseases. The ADMIXTURE program is commonly used to infer individual ancestry from genomic data. Results: We describe two improvements to the ADMIXTURE software. The first enables ADMIXTURE to infer ancestry for a new set of individuals using cluster allele frequencies from a reference set of individuals. Using data from the 1000 Genomes Project, we show that this allows ADMIXTURE to infer ancestry for 10,920 individuals in a few hours (a 5x speedup). This mode also allows ADMIXTURE to correctly estimate individual ancestry and allele frequencies from a set of related individuals. The second modification allows ADMIXTURE to correctly handle X-chromosome (and other haploid) data from both males and females. We demonstrate increased power to detect sex-biased admixture in African-American individuals from the 1000 Genomes project using this extension. Conclusions: These modifications make ADMIXTURE more efficient and versatile, allowing users to extract more information from large genomic datasets.
BibTeX
@misc{shringarpure2016efficient,
  author = {Suyash Shringarpure and Carlos D. Bustamante and Kenneth Lange and David H. Alexander},
  title = {{Efficient analysis of large datasets and sex bias with ADMIXTURE}},
  year = {2016},
  doi = {10.1101/039347},
  howpublished = {bioRxiv},
}
Figure