Leveraging hierarchical population structure in discrete association studies

Area of Science:

• Genomics
• Population Genetics
• Bioinformatics

Background:

• Population structure is a significant confounder in biological data analysis, leading to spurious correlations.
• Existing methods for correcting population structure effects are disparate and discipline-specific.
• Understanding the distinct processes underlying confounding is crucial for accurate biological inference.

Purpose of the Study:

• To examine methods for correcting confounding in discrete data with hierarchical population structure.
• To identify and define two distinct confounding processes: coevolution and conditional influence.
• To develop and apply generative models for correcting these confounding effects in biological data.

Main Methods:

• Development of generative models to describe coevolution and conditional influence processes.
• Application of these models to correct for confounding in biological datasets.
• Evaluation of model performance across diverse biological applications.

Main Results:

• Identified and characterized two distinct confounding processes: coevolution and conditional influence.
• Demonstrated the utility of generative models in correcting for population structure confounding.
• Showcased successful applications in HIV-1 escape mutation identification, peptide coevolution prediction, and Arabidopsis thaliana resistance trait association.

Conclusions:

• No single method is universally optimal for correcting all forms of population structure confounding.
• The choice between coevolution and conditional influence models depends on the specific biological application.
• Generative models provide a flexible framework for addressing confounding in population genetics and related fields.