A complete diploid human genome benchmark for personalized genomics.
Summary
This study introduces a novel, complete telomere-to-telomere diploid human genome benchmark for the HG002 genome. Achieving near-perfect accuracy across 99.4% of the diploid genome, this benchmark incorporates an additional 701.4 Mb of autosomal sequence and 216.8 Mb of sex chromosomes, which were missing from prior benchmarks. The researchers annotated genes and repeats on both haplotypes, identifying nearly 20,000 protein-coding genes on the maternal haplotype and over 19,000 on the paternal. They also developed new methodologies to assess the accuracy of reads, phased variant call sets, and assemblies against this diploid reference. Genome-wide analyses reveal that de novo assembly resolves 2-7% more sequence and significantly outperforms variant calling accuracy.
Analysis
This study marks a significant bioinformatic advancement by providing a complete, high-accuracy, telomere-to-telomere diploid human genome benchmark for HG002. It represents a crucial incremental application of genome assembly methods, specifically designed to overcome technical biases inherent in haploid reference-based mapping approaches that exclude duplicated and structurally polymorphic genomic regions. The benchmark was utilized to compare the accuracy of de novo assembly versus variant calling, demonstrating that de novo assembly is an order of magnitude more accurate and resolves a significantly larger portion of the genome. The training and validation data are based on the HG002 genome, incorporating an additional 701.4 Mb of autosomal sequence and 216.8 Mb of sex chromosomes, along with gene and repeat annotations on both haplotypes. If clinically deployed, this complete diploid reference genome would transform genomic analysis workflows by enabling a more comprehensive and precise characterization of genetic variations, including those in complex regions, which is critical for truly personalized genomic medicine.