Haplotype Phasing

Population-based phasing leverages linkage disequilibrium (LD) —the non-random association of alleles at different loci—to infer haplotypes. Algorithms like SHAPEIT and Eagle use hidden Markov models (HMMs) or Li-Stephens models to estimate the most likely haplotype configuration by referencing large reference panels (e.g., 1000 Genomes Project).

• Mechanism: Iteratively updates phase probabilities based on observed haplotype frequencies in a reference cohort.
• Key advantage: Works with short-read sequencing data where physical linkage is absent.
• Limitation: Accuracy degrades for rare variants or populations underrepresented in reference panels.

Physical phasing directly observes cis-relationships between alleles by using sequencing reads that span multiple heterozygous sites. PacBio HiFi and Oxford Nanopore long reads, or 10x Genomics linked reads, physically connect variants separated by tens of kilobases.

• Mechanism: A single read or read-pair covering two or more heterozygous positions provides unambiguous phase information.
• Key advantage: Resolves phase for rare and de novo variants without population reference panels.
• Metric: Phase block N50 length—the length at which 50% of the phased genome is contained in blocks of that size or larger.

Parental genotyping provides a gold standard for phasing by applying Mendelian transmission constraints. An allele present in a child but absent in one parent must originate from the other parent, resolving phase unambiguously.

• Mechanism: Uses mother-father-child genotype trios to assign alleles to parental haplotypes by identity-by-descent.
• Key advantage: Highest accuracy; resolves phase across entire chromosomes.
• Application: Critical for clinical diagnosis of compound heterozygous disorders where two recessive mutations on opposite haplotypes cause disease.

Phasing algorithms output phase blocks—contiguous genomic segments where the haplotype assignment is internally consistent. The boundary between blocks represents a switch point where the algorithm cannot confidently link variants.

• Switch error rate: The frequency at which the predicted haplotype assignment flips to the opposite parental chromosome. Measured per megabase.
• Benchmarking: Evaluated against trio-phased truth sets or molecular haplotyping data (e.g., Strand-seq, Hi-C phasing).
• Clinical relevance: Long, accurate phase blocks are essential for identifying haploinsufficiency and regulatory haplotype effects.

Hi-C and Strand-seq provide chromosome-scale phasing by physically separating or crosslinking homologous chromosomes before sequencing. Hi-C captures chromatin proximity ligation events that are overwhelmingly intra-chromosomal, while Strand-seq uses BrdU incorporation to selectively sequence single DNA strands.

• Mechanism: Hi-C contacts cluster by chromosome of origin; Strand-seq reads segregate by template strand.
• Key advantage: Achieves whole-chromosome haplotypes without parental data or population references.
• Emerging application: Phasing structural variants and resolving complex genomic rearrangements in cancer genomes.

Accurate phasing is a prerequisite for genotype imputation—the statistical inference of untyped variants from a reference panel. Phased haplotypes are matched against reference haplotypes to fill in missing genotypes, enabling GWAS meta-analysis and polygenic risk score (PRS) calculation across genotyping arrays.

• Workflow: Genotype array → Phasing (Eagle/Beagle) → Imputation (Minimac/IMPUTE) → PRS calculation.
• Impact: Phasing errors propagate to imputation errors, reducing the predictive power of PRS models.
• Scale: Modern biobanks phase and impute hundreds of thousands of samples simultaneously.