Minimizing reference bias with an imputed personalized reference.

Area of Science:

• Genomics
• Bioinformatics

Background:

• Pangenome indexes minimize reference bias in sequencing data.
• Personalized diploid references offer further bias reduction by matching donor alleles.

Purpose of the Study:

• To present a novel impute-first alignment framework combining genotype imputation and read alignment.
• To enhance variant-calling accuracy and reduce errors in whole-genome sequencing.

Main Methods:

• Genotype individuals using a subsample of sequencing reads.
• Impute a personalized diploid reference using a reference panel and imputation algorithm.
• Index the personalized reference and align reads using linear or graph aligners.

Main Results:

• Achieved a higher variant-calling F1 score (99.77%) on HG002 compared to existing methods.
• Demonstrated substantial error reduction: 38.73% vs. linear, 14.97% vs. graph, and 6.05% vs. personalized graph aligners.
• The impute-first approach produced accurate, phased diploid references even from low-coverage data.

Conclusions:

• The impute-first framework provides an advantageous alternative to pangenome references for whole-genome sequencing.
• This approach fully considers linkage disequilibrium and is compatible with both linear and graph reference representations.
• Highest accuracy was achieved using a standard linear aligner (BWA-MEM).