A Bias Correction Method for Substructure Estimation with Small Reference Sample Sizes

Small reference sample sizes for five continental groups (AFR, EAS, EUR, IAM, SAS) introduce systematic bias in Summix2 substructure proportion estimates, with accuracy requiring at least 150 individuals per group.

• Why it matters: Many populations lack large reference panels, and Summix2 relies on reference allele frequencies estimated from finite samples, so reference-sample limitations can distort substructure estimates.
• What they did: The study evaluated Summix2 under simulated Latino-like, African American-like, and Peruvian-like populations plus real African American data, varying reference sample sizes and testing inverse and power-law bias corrections.
• The result: A threshold of 150 per reference group kept mean estimation error below 0.05; power-law correction reduced error most when all groups were small, while inverse correction helped most when only one group was small.

AI summary of the abstract · may contain errors · not medical advice · read the paper.