A rarefaction approach for measuring population differences in rare and common variation

Daniel J. Cotter, E. Hofgard, John Novembre, Zachary A. Szpiech, Noah A. RosenbergGenetics 2023 · DOI

TL;DR

This paper introduces a novel rarefaction-based method to standardize sample sizes when comparing rare and common genetic variations across human populations. This approach helps to disentangle true population differences from artifacts caused by varying sample sizes, leading to more robust insights into human genetic diversity.

What Problem Does This Paper Address?

When studying genetic variation across different human populations, researchers often categorize alleles as "rare" (low frequency), "common" (high frequency), or "unobserved" within a population. However, a significant challenge arises when these populations have different numbers of individuals sampled. A small sample size might miss alleles that are actually present but rare, making that population appear to have less variation than a population with a larger sample size, even if their underlying genetic makeup is similar. This discrepancy can lead to misinterpretations of population differences. The paper aims to address this by developing a statistical method to correct for these sample size differences, allowing for a more accurate comparison of rare and common genetic variation across populations.

Methods

The core of the paper's contribution is a rarefaction-based sample-size correction method. This approach, adapted from ecological studies of species diversity, involves calculating the expected number of rare and common alleles in hypothetical subsamples of equal size drawn from each population.

Here's a breakdown of the key methodological aspects:

  • Three Allelic Classes: The method explicitly defines three categories for an allelic type within a population:
  • Unobserved (U): Not detected in the sample.
  • Rare (R): Detected, but with a frequency below a specified threshold (e.g., 5%).
  • Common (C): Detected, with a frequency above the specified threshold.
  • Probabilistic Framework: For each allelic type and each population, the method calculates the probabilities of it falling into each of the U, R, or C categories, given a specific subsample size g. This involves combinatorial formulas to account for the number of ways an allele can be observed or not observed in a subsample.
  • Generalization to Multiple Populations: The probabilities for individual populations are then combined to determine the probability of a specific pattern of U, R, and C across all populations. This allows for the comparison of complex geographic distributions of genetic variation.
  • Extension to More Than Three Classes: The framework is flexible and can be extended to define more than three frequency classes, allowing for finer-grained distinctions of rare variation.
  • Sliding-Window Analysis: The method is applied to analyze genetic variation along the genome by calculating pattern probabilities within sliding windows, enabling the identification of genomic regions with distinct patterns of rare and common variation.

The novelty lies in extending the rarefaction concept, previously used for presence/absence (unobserved vs. observed), to differentiate between rare and common variation, and applying it to human population genetics.

Datasets

The primary dataset used for analysis is from the 1000 Genomes Project. Specifically, the authors utilized data from the 2,504 individuals across 26 populations, which were grouped into five "super-populations" (African, European, South Asian, East Asian, and American) as defined in Biddanda et al. (2020).

The analysis focused on biallelic Single Nucleotide Polymorphisms (SNPs). For some analyses, "nonsingleton" SNPs were considered, meaning SNPs where the minor allele appeared more than once in the dataset, to reduce the influence of very rare variants.

Key Insights and Ideas

  • Sample Size Matters (Even When You Think It Doesn't): The most crucial insight is that even with relatively large sample sizes (like those in the 1000 Genomes Project), differences in sample size can subtly influence the classification of alleles as rare or common. This is due to "discreteness effects" – the chance of observing a specific number of copies of an allele in a sample is sensitive to sample size, especially for alleles near the rare/common frequency threshold.
  • Rare Variation is Often Population-Specific: The analysis reinforces the understanding that much of human genetic variation is rare and often specific to particular populations or groups of populations. The rarefaction method helps to quantify this more accurately by accounting for sample size.
  • Common Variation is Truly Common: Conversely, alleles that are truly common tend to be common across most, if not all, human populations. The method helps to distinguish these from alleles that are only classified as common due to sampling artifacts.
  • Visualizing Genetic Differences: The paper emphasizes the power of visualization, building upon previous work, to illustrate complex patterns of genetic variation. The "Geovar" plots and their extensions provide intuitive ways to grasp population-level genetic differences.

Key Results

  • Subtle but Meaningful Differences: The sample-size correction introduced by the authors leads to subtle differences in the probabilities of various geographic patterns of rare and common variation compared to analyses that use full sample sizes. These differences are more pronounced at smaller subsample sizes (g).
  • Agreement at Larger Subsample Sizes: As the subsample size (g) increases (e.g., to 500), the sample-size-corrected probabilities closely match the empirical pattern frequencies observed in the full dataset, suggesting that the correction is effective.
  • Impact of Singletons: Excluding singleton variants (those appearing only once) significantly alters the landscape of observed patterns, revealing more complex distributions of rare variation that were previously masked by the overwhelming presence of singletons.
  • Genomic Location Matters: Sliding-window analyses reveal that while overall patterns of rare and common variation are broadly consistent across the genome, specific regions (like the HLA region) show deviations, potentially reflecting localized evolutionary pressures such as balancing selection.

Limitations and Future Work

  • Choice of Threshold: The definition of "rare" and "common" depends on a chosen frequency threshold (e.g., 5%). The paper acknowledges that the optimal threshold might vary depending on the specific sample sizes and the underlying allele frequency distributions.
  • Data Quality: The authors note that data quality can vary across genomic regions, which could affect the accuracy of pattern probability measurements, particularly in regions with high variation or potential technical artifacts.
  • Future Directions:
    • Ancient and Modern Data: The method is well-suited for comparing ancient DNA samples (often with very small sample sizes) with modern populations.
  • Finer-Grained Rare Variation: The framework can be extended to define multiple tiers of rare variation, allowing for more detailed studies of low-frequency alleles.
  • Application to Other Organisms: The approach could be adapted for studying population structure and variation in non-human species.
  • Understanding Evolutionary Processes: The sliding-window analysis provides a foundation for using these pattern probabilities to infer demographic history and selection pressures.

Summary updated

Abstract

In studying allele-frequency variation across populations, it is often convenient to classify an allelic type as "rare," with nonzero frequency less than or equal to a specified threshold, "common," with a frequency above the threshold, or entirely unobserved in a population. When sample sizes differ across populations, however, especially if the threshold separating "rare" and "common" corresponds to a small number of observed copies of an allelic type, discreteness effects can lead a sample from one population to possess substantially more rare allelic types than a sample from another population, even if the two populations have extremely similar underlying allele-frequency distributions across loci. We introduce a rarefaction-based sample-size correction for use in comparing rare and common variation across multiple populations whose sample sizes potentially differ. We use our approach to examine rare and common variation in worldwide human populations, finding that the sample-size correction introduces subtle differences relative to analyses that use the full available sample sizes. We introduce several ways in which the rarefaction approach can be applied: we explore the dependence of allele classifications on subsample sizes, we permit more than two classes of allelic types of nonzero frequency, and we analyze rare and common variation in sliding windows along the genome. The results can assist in clarifying similarities and differences in allele-frequency patterns across populations.
BibTeX
@inproceedings{cotter2023rarefaction,
  author = {Daniel J. Cotter and E. Hofgard and John Novembre and Zachary A. Szpiech and Noah A. Rosenberg},
  title = {{A rarefaction approach for measuring population differences in rare and common variation}},
  booktitle = {Genetics},
  year = {2023},
  doi = {10.1093/genetics/iyad070},
}
Figure