EMU

EMU infers population structure from large-scale genetic datasets with substantial missing genotype data.

Key Features:

• Handling Missing Data: Manages non-random missingness in genotype data to prevent bias in principal component analyses.
• EM-PCA algorithm: Implements an Expectation-Maximization Principal Component Analysis (EM-PCA) approach to infer population structure despite rampant missingness.
• Performance and scalability: Shows superior accuracy and competitive computational efficiency relative to alternative PCA methods based on extensive simulations.
• Large-cohort capability: Recovered Han Chinese population structure in the Chinese Millionome Project Phase 1 dataset (~100,000 individuals at ~0.08× coverage) using CPU hours.

Scientific Applications:

• Ultra-low coverage sequencing: Infer population structure from datasets generated by non-invasive prenatal tests and other ultra-low coverage sequencing protocols.
• Shallowly sequenced cohorts: Analyze population structure in large shallow-sequenced projects such as the Chinese Millionome Project Phase 1.
• Population-genetics analyses: Provide population structure estimates required for downstream genetic and genomic analyses that must account for missing genotype data.

Methodology:

Uses an Expectation-Maximization Principal Component Analysis (EM-PCA) algorithm and evaluation via extensive simulations comparing to other PCA methods.