Clade distillation for genome-wide association studies.

Area of Science:

• Human genetics
• Population genetics
• Statistical genetics

Background:

• Testing inferred haplotype genealogies for phenotype association is a key goal in human genetics.
• Detecting association signals driven by allelic heterogeneity (multiple causal variants) in coding and noncoding regions is challenging.
• Scalable methods for inferring genealogical trees exist, but testing these trees for association remains unsolved due to computational complexity.

Purpose of the Study:

• To introduce practical improvements to the kalis ancestry inference engine for efficient genome-wide analyses.
• To propose LOCATER, a novel procedure for testing local tree representations for trait association.
• To enable genealogy-based association testing for any ancestry inference engine output.

Main Methods:

• Implemented optimal checkpointing for decoding hidden Markov models in the kalis engine.
• Developed LOCATER, a procedure utilizing the Stable Distillation framework for tree-based association testing.
• Evaluated LOCATER using simulated quantitative phenotypes and compared it to traditional methods.

Main Results:

• LOCATER demonstrates substantial power gains over single marker and window-based testing for allelic heterogeneity.
• The method improves causal region localization compared to existing approaches.
• LOCATER shows significant advantages in detecting association signals driven by multiple ultra-rare variants.

Conclusions:

• Genealogy-based association testing is a powerful approach for gene discovery.
• LOCATER offers a robust and flexible method for phenotype-genotype association studies.
• This work advances the field of human genetics by providing tools to unravel complex genetic architectures.