Summix

Summix estimates ancestry proportions from genetic summary statistics and adjusts allele frequencies to correct for population structure and reduce confounding in genetic analyses.

Key Features:

• Ancestry Deconvolution: Estimates ancestry proportions in summary statistics using continental reference ancestries African (AFR), Non-Finnish European (EUR), East Asian (EAS), Indigenous American (IAM), and South Asian (SAS), with reported precision within 0.1% in simulations.
• Ancestry-Adjusted Allele Frequencies: Adjusts allele frequencies for population structure and aligns adjusted frequencies to reference samples, with demonstrated application to gnomAD v2.1 exome and genome groups.
• Efficiency and Speed: Processes dense panels of summary statistics in seconds and supports estimation of confidence intervals via block bootstrap methods.
• Improved Representation: Produces ancestry-adjusted summary data to improve representation and comparability for understudied or admixed populations.
• Implementation: Implemented as an R package supporting analyses such as prioritizing putative causal variants, computing polygenic scores, and leveraging common controls.

Scientific Applications:

• Reducing Confounding: Reduces confounding by population structure to improve identification of true genetic associations.
• Variant Prioritization: Prioritizes putative causal variants by providing population-structure-adjusted allele frequencies.
• Polygenic Scoring: Improves construction and transferability of polygenic risk scores by adjusting allele frequencies for ancestry.
• Common Controls: Enables use of public datasets as common controls by adjusting allele frequencies to match target ancestry compositions.
• Admixed and Understudied Populations: Supports analyses in admixed and understudied populations by estimating ancestry proportions from summary-level data.

Methodology:

Estimates ancestry proportions in summary statistics using continental reference ancestries (AFR, EUR, EAS, IAM, SAS), adjusts allele frequencies to reference compositions, and computes confidence intervals via block bootstrap; methods demonstrated on gnomAD v2.1 exome and genome groups.