# Haplotype-Resolved Assembly with Long Reads: How to Separate Maternal and Paternal Genomes

Researchers studying allele-specific expression, compound heterozygosity, or structural variation need genome assemblies that preserve the distinction between the two parental chromosome copies. Standard diploid assembly collapses homologous sequences into a single consensus, hiding which variants travel together on the same chromosome. Haplotype-resolved assembly with long reads solves this problem by producing two separate sequence representations, one for each parental genome. This article explains the practical workflow for generating phased assemblies using HiCanu, Falcon-Unzip, and hifiasm, with strategies for incorporating parental data or Hi-C chromatin information when trio samples are unavailable.

## The Core Problem: Why Diploid Collapse Hides Biological Information

Most genome assemblers were designed to produce a single reference sequence from a sample that contains two distinct chromosome copies. When the two parental copies differ by enough sequence variation, the assembler must decide which variant to represent in the final output. This process, called collapse, merges the two haplotypes into one mosaic sequence that does not exist in nature. The result is a reference that contains some variants from the mother and some from the father, with no information about which combinations occur together on the same physical chromosome.

The biological consequences of collapse are substantial. For a gene with two deleterious variants in different exons, one on each parental chromosome, a collapsed assembly shows both variants in a single sequence. This creates the false impression that one chromosome carries both mutations. In reality, the patient is a compound heterozygote, with each parent contributing one variant. The distinction matters for diagnosis, genetic counseling, and understanding disease mechanism. Long-read sequencing technologies now permit routine analysis of DNA fragments tens to hundreds of kilobase pairs in size, which enables the direct observation of variants that are physically linked on the same molecule [7]. This physical linkage is the foundation of haplotype-resolved assembly.

The practical question for a laboratory is not whether phasing is valuable but which combination of sequencing technology, assembler, and additional data types will produce the most complete and accurate separation of parental genomes for a given budget and sample type.

## At a Glance: Assembler and Data Strategy Comparison

| Assembler | Input Data Requirements | Phasing Strategy | Best Use Case | Key Limitation |
|-----------|------------------------|------------------|---------------|----------------|
| HiCanu | PacBio continuous long reads or Oxford Nanopore reads | Overlap-based assembly with haplotype-aware graph construction | Large genomes with moderate heterozygosity where graph outputs are needed for downstream phasing | Requires high coverage, produces primary and alternate contigs that need additional phasing steps |
| Falcon-Unzip | PacBio continuous long reads | Primary contig assembly followed by haplotype splitting using read information | Diploid genomes with sufficient heterozygosity to distinguish haplotypes | Works best with PacBio data, performance depends on variant density between haplotypes |
| hifiasm | PacBio HiFi reads (highly accurate circular consensus reads) | Graph-based assembly with integrated phasing using Hi-C, parental data, or both | Human and other mammalian genomes where HiFi data are available | Requires HiFi sequencing, which has shorter read lengths than continuous long-read approaches |
| FALCON-Phase | Hi-C reads plus Falcon or Falcon-Unzip draft assembly | Uses chromatin interaction data to extend phase blocks to chromosome scale | Samples where parental DNA is unavailable | Accuracy improves with higher heterozygosity, lower accuracy in low-diversity regions |

The choice among these approaches depends on the sequencing platform available, whether parental samples can be obtained, and the level of heterozygosity in the target species. The following sections detail how each strategy works and how to integrate them into a reproducible analysis pipeline.

## Long-Read Sequencing Platforms and Their Impact on Phasing

### PacBio Continuous Long Reads and HiFi Reads

Pacific Biosciences sequencing produces two distinct data types that serve different assembly purposes. Continuous long reads are generated by the original sequencing-by-synthesis approach and provide very long molecules but with lower per-base accuracy. The optimization of circular consensus sequencing improved the accuracy of single-molecule real-time sequencing to generate highly accurate long high-fidelity reads with an average length of 13.5 kilobases [10]. These HiFi reads achieve approximately 99.8% accuracy, which means they can be used both for accurate variant detection and for assembly.

The accuracy difference matters for phasing. Continuous long reads contain enough errors that the assembler must use consensus approaches to distinguish true variation from sequencing noise. HiFi reads have sufficient accuracy that individual reads can be trusted for variant detection, which simplifies the assembly graph and makes haplotype separation more straightforward. The original HiFi demonstration on the human HG002 genome showed that nearly all variants could be phased into haplotypes, with de novo assembly producing a contig N50 above 15 megabases and concordance of 99.997% [10].

### Oxford Nanopore Sequencing

Oxford Nanopore technology offers an alternative long-read platform with different tradeoffs. Nanopore sequencing permits direct electronic detection of DNA molecules and can produce very long reads, often exceeding 100 kilobases. The error profile differs from PacBio, with a higher proportion of insertion and deletion errors in early versions of the technology. Canu was specifically designed to handle noisy single-molecule sequences from both PacBio and Oxford Nanopore platforms, introducing support for nanopore data and reducing depth-of-coverage requirements [8].

For clinical applications, nanopore sequencing has demonstrated practical utility in rare disease diagnosis. A study of 98 samples from 41 families with suspected rare monogenic diseases used nanopore sequencing at approximately 36-fold average coverage with a 32-kilobase read N50 from a single flow cell per sample [9]. This study showed that long-read sequencing could phase 87% of protein-coding genes completely and detect additional structural variants and tandem repeats that short-read sequencing missed [9]. The diagnostic yield improvement came from the ability to phase variants and distinguish compound heterozygotes from simple heterozygotes.

### Read Length and Coverage Requirements

The physical length of sequencing reads determines how far apart two variants can be and still be observed on the same molecule. If a read spans both variants, the phase is directly observed. If the variants are farther apart than the read length, the phase must be inferred through assembly graphs or additional data types such as Hi-C.

Coverage requirements depend on the assembly strategy. Overlap-based assemblers like Canu need sufficient depth to distinguish true overlaps from repetitive sequence matches. The Canu publication demonstrated that the assembler could reliably assemble complete microbial genomes and near-complete eukaryotic chromosomes using either PacBio or Oxford Nanopore technologies, achieving a contig NG50 above 21 megabases on human data [8]. The key algorithmic advance was an adaptive overlapping strategy based on tf-idf weighted MinHash, which reduced the computational cost while improving repeat separation [8].

For HiFi-based assembly with hifiasm, coverage of 30 to 40-fold is typically recommended for diploid genomes. The high accuracy of HiFi reads means that lower coverage can be used compared to continuous long-read approaches, but sufficient depth is still needed to distinguish sequencing errors from true heterozygosity.

## Core Principles of Haplotype Separation

### The Assembly Graph as the Foundation

All haplotype-resolved assemblers operate on a graph representation of the sequencing reads. The graph contains nodes that represent sequence segments and edges that represent adjacencies supported by read evidence. In a diploid genome, regions where the two haplotypes differ create bubbles in the graph, with two alternative paths representing the maternal and paternal sequences.

The challenge is distinguishing true haplotype variation from sequencing errors. In a perfect world with error-free reads, every heterozygous variant would create a clean bubble in the assembly graph. With real sequencing data, errors create spurious bubbles and complicate the graph structure. The assembler must use coverage, base quality, and read context to determine which bubbles represent genuine haplotypes and which represent noise.

Canu addressed this challenge with a sparse assembly graph construction that avoids collapsing diverged repeats and haplotypes [8]. The assembler produces graph-based assembly outputs in graphical fragment assembly format for analysis or integration with complementary phasing and scaffolding techniques [8]. This graph output is the key intermediate that enables downstream phasing tools to extend phase blocks.

### Primary and Alternate Contigs

A common strategy in diploid assembly is to produce a primary assembly that represents one haplotype and a set of alternate contigs that represent the other haplotype in regions where the two differ. Falcon-Unzip uses this approach, first assembling the genome into primary contigs and then using read information to split the primary contigs into haplotype-specific sequences.

The primary contig represents the more common or better-supported allele at each position, while the alternate contigs capture the alternative allele. This representation is useful for many applications because the primary assembly provides a complete reference while the alternates preserve haplotype information. However, the primary and alternate representation does not produce two complete chromosome-length haplotypes. Instead, it produces a mosaic primary sequence with alternate segments that need to be assigned to one haplotype or the other.

### Phase Blocks and Their Extension

A phase block is a genomic region where the phase of variants is known. Within a phase block, all variants are assigned to one of two haplotypes, and the assignment is consistent across the block. The boundaries of phase blocks occur where the evidence for phase assignment becomes ambiguous, often in regions of low heterozygosity or high repetitiveness.

Long reads create phase blocks that span the length of individual reads or read pairs. For a 13.5-kilobase HiFi read, the phase block extends across that read length. For continuous long reads of 100 kilobases or more, the phase block is correspondingly longer. The assembly process connects these read-level phase blocks into longer blocks through the assembly graph.

Hi-C data can extend phase blocks to chromosome scale. FALCON-Phase uses ultra-long-range Hi-C chromatin interaction data to extend phase blocks of partially phased diploid assemblies to chromosome or scaffold scale [11]. The method uses the inherent phasing information in Hi-C reads, skipping variant calling and reducing computational complexity [11]. Validation on human, cow, and zebra finch genomes showed accuracy of 97% for cow and zebra finch compared to 80 to 91% for human, with better performance in samples with higher heterozygosity [11].

## Assembler-Specific Workflows

### HiCanu: Graph-Based Assembly for Noisy Long Reads

HiCanu is the haplotype-aware version of the Canu assembler. It is designed for continuous long reads from PacBio or Oxford Nanopore platforms and produces assembly graphs that preserve haplotype structure. The workflow begins with read correction, where the assembler uses read-to-read overlaps to identify and correct sequencing errors. This correction step is computationally intensive but essential for downstream assembly quality.

The corrected reads are then assembled into unitigs, which are sequences that are supported by consistent read overlaps. In diploid regions, the unitig graph contains bubbles that represent the two haplotypes. HiCanu traverses these bubbles to produce primary contigs and alternate contigs, with the graph structure preserved in the output.

The key decision for HiCanu users is coverage. The assembler halves depth-of-coverage requirements compared to its predecessor Celera Assembler, but sufficient coverage is still needed for the overlap step to work reliably [8]. For human genomes, coverage of 50 to 60-fold with continuous long reads is a common starting point. Lower coverage may work for smaller genomes or genomes with higher heterozygosity, where haplotype separation is easier.

The output of HiCanu includes both the assembled contigs and the assembly graph in graphical fragment assembly format [8]. This graph output is valuable for downstream analysis because it preserves the haplotype structure that is lost in a simple fasta file. Tools like FALCON-Phase can use this graph information to extend phase blocks.

### Falcon-Unzip: Primary Contig Splitting

Falcon-Unzip follows a two-stage approach. The first stage, Falcon, assembles the genome into primary contigs using a string graph approach. The second stage, Unzip, uses the read alignments to the primary contigs to identify heterozygous variants and split the contigs into haplotype-specific segments.

The Unzip step works by aligning all reads back to the primary contigs and identifying positions where reads support two different alleles. These positions mark the boundaries of haplotype-specific segments. The segments are then assigned to haplotypes based on the pattern of allele support across the segment.

Falcon-Unzip works best when the two haplotypes differ sufficiently that reads can be unambiguously assigned to one haplotype or the other. In regions of low heterozygosity, the assignment becomes ambiguous and the assembler may produce collapsed sequence. The heterozygosity threshold depends on the read length and error rate, but in practice, species with heterozygosity below approximately 0.1% present challenges for this approach.

The output of Falcon-Unzip is a set of primary contigs and a set of alternate contigs (the unzipped haplotypes). The primary contigs provide a complete reference, while the alternate contigs provide the alternative haplotype in regions where it differs. This output format is compatible with FALCON-Phase, which can use the primary and alternate contigs to produce chromosome-scale haplotypes [11].

### hifiasm: Integrated Phasing with HiFi Reads

hifiasm is designed specifically for HiFi reads and integrates phasing into the assembly process instead of treating it as a post-assembly step. The assembler builds a graph from the HiFi reads and uses the high accuracy of the reads to distinguish true haplotypes from sequencing errors.

The key advantage of hifiasm is that it can use additional data types to improve phasing. When parental data are available, hifiasm can use the parental reads or variants to assign contigs to maternal and paternal haplotypes. When Hi-C data are available, hifiasm can use the chromatin interaction information to extend phase blocks and produce chromosome-scale haplotypes.

The hifiasm workflow is straightforward from the user perspective. The input is a set of HiFi reads in fastq format, optionally with parental reads or Hi-C reads. The output is a set of haplotype-resolved contigs, with maternal and paternal haplotypes represented as separate sequences.

The quality of hifiasm assemblies depends on the accuracy of the HiFi reads and the depth of coverage. The original HiFi demonstration showed that circular consensus sequencing could achieve 99.8% accuracy with an average read length of 13.5 kilobases [10]. This accuracy level means that individual reads can be trusted for variant detection, which simplifies the assembly graph and improves haplotype separation.

### FALCON-Phase: Hi-C Extension Without Parental Data

FALCON-Phase addresses the common situation where parental samples are unavailable. The method uses Hi-C chromatin interaction data to extend phase blocks of partially phased diploid assemblies to chromosome or scaffold scale [11]. The approach skips variant calling entirely, using the inherent phasing information in Hi-C reads to assign contigs to haplotypes [11].

The input to FALCON-Phase is a draft assembly that contains long primary contigs and phased alternate contigs, plus Hi-C reads aligned to the assembly. The method works by examining the pattern of Hi-C interactions between primary and alternate contigs. Contigs that show more Hi-C interactions with each other than expected by chance are assigned to the same haplotype.

The accuracy of FALCON-Phase depends on the heterozygosity of the sample. Validation on three benchmark datasets from the Vertebrate Genomes Project showed accuracy of 97% for cow and zebra finch compared to 80 to 91% for human [11]. The lower accuracy in human reflects the lower heterozygosity of human genomes, which provides less signal for haplotype assignment.

The practical implication is that FALCON-Phase is most useful for species with moderate to high heterozygosity. For low-heterozygosity species, the accuracy may be insufficient for clinical or diagnostic applications, and trio-based approaches should be preferred when possible.

## Using Parental Data for Trio Phasing

### The Trio Strategy

When DNA from both parents is available, trio phasing provides the most accurate assignment of haplotypes. The strategy relies on the principle that each child inherits one chromosome from each parent. By comparing the child's reads to the parental genotypes, each read can be assigned to the maternal or paternal haplotype based on the variants it carries.

The trio approach requires sequencing or genotyping of both parents. The parental data do not need to be as deep as the child's data, since the parents are only used to establish the phase of variants, not to assemble their genomes. In practice, 10 to 20-fold coverage of each parent with short reads is often sufficient for variant detection, which can then be used to phase the child's long reads.

The advantage of trio phasing is that it provides a definitive assignment of haplotypes. There is no ambiguity about which haplotype came from which parent, which matters for studies of parent-of-origin effects and for clinical applications where the inheritance pattern is relevant.

### Integration with Assemblers

Both hifiasm and HiCanu can use parental data to improve phasing. In hifiasm, the parental reads or variants are provided as additional input, and the assembler uses them to assign contigs to haplotypes during the assembly process. In HiCanu, the parental data can be used after assembly to assign primary and alternate contigs to parental haplotypes.

The integration of parental data is most valuable in regions of low heterozygosity, where the assembly graph does not provide enough signal to distinguish haplotypes. In these regions, the parental genotypes can resolve the phase even when the child's reads do not span both variants.

The limitation of trio phasing is the requirement for parental samples. For many species, including most non-model organisms, parental samples are not available. For human clinical applications, parental samples may be available but add cost and logistical complexity. The decision to use trio phasing should weigh the value of definitive haplotype assignment against the cost of additional sequencing.

## Hi-C Data as an Alternative to Parental Samples

### How Hi-C Provides Phasing Information

Hi-C captures the three-dimensional structure of chromatin by crosslinking DNA that is in close physical proximity in the nucleus. After sequencing, the resulting read pairs connect genomic regions that are near each other in three-dimensional space. Because the two homologous chromosomes are physically separated in the nucleus, Hi-C read pairs predominantly connect regions on the same chromosome copy.

This physical separation provides phasing information. If a Hi-C read pair connects a region on the primary contig to a region on an alternate contig, the two contigs are likely on the same haplotype. By accumulating these interaction patterns across many read pairs, the phase of contigs can be determined.

FALCON-Phase exploits this principle by examining Hi-C interaction patterns between primary and alternate contigs [11]. The method skips variant calling, which reduces computational complexity and makes it applicable to any draft assembly that contains long primary contigs and phased alternate contigs [11].

### Practical Considerations for Hi-C Phasing

The quality of Hi-C phasing depends on the depth of Hi-C sequencing and the quality of the draft assembly. Higher Hi-C coverage provides more interaction data and more confident phase assignments. The draft assembly must contain phased alternate contigs, which means that the initial assembly must have successfully separated the haplotypes in at least some regions.

The accuracy of Hi-C phasing varies with heterozygosity. In the Vertebrate Genomes Project validation, accuracy was higher for cow and zebra finch than for human, reflecting the higher heterozygosity of those species [11]. For low-heterozygosity species, the Hi-C signal may be insufficient to distinguish haplotypes, and the phase assignments may be incorrect in a substantial fraction of the genome.

Hi-C phasing is most useful when parental samples are unavailable and when the species has sufficient heterozygosity to support haplotype separation in the initial assembly. For species with very low heterozygosity, the initial assembly may not produce phased alternate contigs, and Hi-C phasing cannot compensate for this limitation.

## Practical Workflow for Haplotype-Resolved Assembly

### Step 1: Assess Sample and Data Requirements

Before beginning an assembly project, evaluate the biological question and the available samples. If the goal is to study allele-specific expression or compound heterozygosity, definitive haplotype assignment is essential, and trio phasing should be preferred when parental samples are available. If the goal is to produce a reference-quality assembly for a species with no prior reference, Hi-C phasing may be sufficient.

The sequencing platform decision follows from the assembly strategy. HiFi reads provide the best combination of accuracy and length for haplotype-resolved assembly and are the preferred input for hifiasm. Continuous long reads from PacBio or Oxford Nanopore can be used with HiCanu or Falcon-Unzip but require higher coverage to achieve comparable results.

### Step 2: Generate Sequencing Data

For HiFi-based assembly, generate at least 30 to 40-fold coverage of the target genome with circular consensus sequencing. The average read length should be above 10 kilobases to provide sufficient phasing information. For continuous long-read assembly, generate 50 to 60-fold coverage with reads as long as possible.

If parental data will be used for trio phasing, generate 10 to 20-fold coverage of each parent with short reads or low-pass long reads. The parental data only need to be sufficient for variant detection, not assembly.

If Hi-C will be used for phasing, generate at least 30-fold coverage of Hi-C data. The Hi-C library preparation requires fresh or properly preserved tissue, since the crosslinking step must occur in intact nuclei.

### Step 3: Run the Assembler

For HiFi reads, run hifiasm with the appropriate options for the data types available. If parental data are available, provide the parental reads or variants as input. If Hi-C data are available, provide the Hi-C reads as input. The assembler will produce haplotype-resolved contigs with the phasing information integrated into the assembly.

For continuous long reads, run HiCanu or Falcon-Unzip. HiCanu produces primary and alternate contigs plus an assembly graph. Falcon-Unzip produces primary contigs and unzipped haplotypes. Both outputs can be used with FALCON-Phase for Hi-C-based phasing.

### Step 4: Evaluate Assembly Quality

Assess the completeness of the assembly using standard metrics such as contig N50, total assembly size, and completeness of expected gene content. For haplotype-resolved assemblies, also assess the completeness of each haplotype separately. The two haplotypes should be similar in size and completeness, with differences reflecting genuine heterozygosity instead of assembly artifacts.

Check the phasing accuracy by examining known heterozygous variants. If parental data are available, verify that the haplotypes match the expected parental inheritance pattern. If Hi-C data were used, check that the phase blocks extend to chromosome scale and that the assignment is consistent across the genome.

### Step 5: Document and Archive

Record the assembly parameters, data versions, and quality metrics in a reproducible format. The nf-core documentation provides standards for community pipeline usage and configuration that can be adapted for documenting assembly workflows [5]. The Galaxy Training Network offers accessible workflow training that emphasizes reproducibility in analysis [4].

Archive the raw sequencing data, the assembly intermediate files, and the final assembly in appropriate repositories. The NCBI provides databases and search systems for sequence data and assembly resources [1]. The EMBL-EBI training resources provide guidance on data submission and analysis best practices [2].

## Records and Measurements for Assembly Projects

### Essential Quality Metrics

| Metric | What It Measures | Target Range | Action If Out of Range |
|--------|------------------|--------------|------------------------|
| Contig N50 | Length at which half the assembly is in contigs of this size or larger | Above 10 Mb for mammalian genomes with HiFi data | Increase coverage or adjust assembler parameters |
| Haplotype completeness | Proportion of expected genes present in each haplotype | Above 90% for each haplotype | Check for collapse in low-heterozygosity regions |
| Phase block N50 | Length at which half the phased genome is in phase blocks of this size or larger | Chromosome scale with Hi-C or trio data | Add Hi-C data or parental data to extend phase blocks |
| Base accuracy | Consensus accuracy of the assembly | Above 99.9% for HiFi-based assembly | Check for systematic errors in homopolymer or repeat regions |
| Variant concordance | Agreement between assembly variants and independently detected variants | Above 99% for single-nucleotide variants | Investigate regions of discordance for assembly errors |

### Recording Requirements

Maintain a laboratory notebook or electronic record that documents the sample identity, DNA extraction method, sequencing platform and version, coverage achieved, assembler version and parameters, and quality metrics for each assembly attempt. This record enables troubleshooting when assemblies fail and provides the provenance needed for publication or clinical reporting.

The Carpentries lessons provide foundational training in data organization and reproducible computing that is directly applicable to managing assembly projects [6]. The Bioconductor project offers official documentation for reproducible genomic analysis workflows that can be adapted for assembly quality assessment [3].

## Common Failure Patterns and Troubleshooting

### Haplotype Collapse in Low-Heterozygosity Regions

The most common failure in haplotype-resolved assembly is collapse of the two haplotypes into a single sequence in regions where the haplotypes are too similar to distinguish. This failure manifests as a reduction in total assembly size, with the assembly being closer to one genome size than two. The collapsed regions lose heterozygosity information and may contain mosaic sequences that combine variants from both haplotypes.

Troubleshooting collapse requires identifying the affected regions and understanding why the assembler could not separate them. Low heterozygosity, high repetitiveness, and insufficient coverage are the usual causes. Increasing coverage may help, but in regions of very low heterozygosity, no amount of coverage will enable separation. In these cases, parental data or Hi-C data may be needed to resolve the phase.

### Misassembly at Repeats

Repetitive regions create ambiguity in the assembly graph because reads from different copies of a repeat appear identical. The assembler may incorrectly join sequences from different genomic locations, creating chimeric contigs. These misassemblies are particularly problematic for haplotype-resolved assembly because they can mix sequences from the two haplotypes.

The Canu approach to repeat separation uses adaptive k-mer weighting to distinguish true overlaps from spurious matches in repetitive regions [8]. The sparse assembly graph construction avoids collapsing diverged repeats and haplotypes [8]. Despite these advances, highly repetitive regions such as centromeres and acrocentric short arms remain challenging, and the complete assembly of these regions required specialized approaches [7].

### Phase Switch Errors

Phase switch errors occur when the assignment of variants to haplotypes is correct within a phase block but the orientation of the block is flipped relative to the true haplotypes. These errors are invisible in most quality metrics because the assembly is internally consistent. They only become apparent when comparing to parental genotypes or when examining the inheritance pattern of variants.

Detection of phase switch errors requires external validation data. If parental genotypes are available, the haplotypes can be checked for Mendelian consistency. If Hi-C data were used for phasing, the interaction patterns can be examined for evidence of incorrect assignment. The accuracy of Hi-C phasing varies by species, with lower accuracy in low-heterozygosity samples [11].

### Insufficient Coverage

Insufficient coverage leads to fragmented assemblies with short contigs and incomplete haplotype separation. The assembler may produce many small contigs that cannot be assigned to haplotypes, resulting in a highly fragmented representation of the genome.

The coverage requirement depends on the sequencing platform and assembler. HiCanu halves depth-of-coverage requirements compared to Celera Assembler, but still needs sufficient depth for reliable overlap detection [8]. HiFi-based assembly with hifiasm requires less coverage than continuous long-read assembly because the higher accuracy simplifies the graph.

## Limitations and Interpretation Boundaries

### What Haplotype-Resolved Assembly Cannot Provide

A haplotype-resolved assembly provides the sequence of the two parental chromosomes but does not directly provide information about gene expression, methylation, or chromatin state. These functional data require additional assays. The assembly is a static representation of the genome, while the functional state of the genome is dynamic and cell-type specific.

The assembly also cannot resolve all regions of the genome. Highly repetitive regions, including centromeres and acrocentric short arms, remained intractable until recently, and their complete assembly required specialized approaches [7]. Even with long-read sequencing, some regions may remain unresolved due to extreme repetitiveness or insufficient coverage.

### Accuracy Limitations in Low-Complexity Regions

The accuracy of haplotype-resolved assembly varies across the genome. Low-complexity regions, including homopolymers and simple repeats, are prone to assembly errors even with high-accuracy HiFi reads. The original HiFi demonstration showed high accuracy for single-nucleotide variants and small indels, but the accuracy in repetitive regions was lower [10].

For clinical applications, the accuracy limitations in low-complexity regions have practical implications. Variants in these regions may be missed or incorrectly assembled, leading to false-negative or false-positive results. The rare disease study using nanopore sequencing showed that long-read sequencing could detect additional variants in regions not covered by short-read sequencing, but the study also acknowledged the limitations of the approach [9].

### Heterozygosity Dependence of Phasing Accuracy

The accuracy of phasing depends on the heterozygosity of the sample. Higher heterozygosity provides more signal for distinguishing haplotypes, while lower heterozygosity makes the distinction more difficult. This dependence is particularly pronounced for Hi-C-based phasing, where the accuracy for human was 80 to 91% compared to 97% for cow and zebra finch [11].

The practical implication is that phasing accuracy should be validated for each species and sample type. A phasing strategy that works well for a highly heterozygous species may be inadequate for a low-heterozygosity species. The validation should include comparison to independently determined haplotypes, such as those from trio data or from experimentally separated haplotypes.

## Quality Control and Validation Strategies

### Internal Consistency Checks

Several quality checks can be performed without external data. The two haplotypes should be similar in size and completeness, with differences reflecting genuine heterozygosity. The number of heterozygous variants should be consistent with the expected heterozygosity of the species. The assembly graph should be examined for unresolved bubbles or other anomalies.

The completeness of each haplotype can be assessed using expected gene content. If the species has a well-annotated reference or a closely related species with a reference, the expected gene set can be compared to the genes present in each haplotype. Missing genes may indicate collapse or misassembly.

### External Validation with Independent Data

The most rigorous validation uses independent data that were not used in the assembly. Parental genotypes provide the gold standard for phasing validation, since the inheritance pattern is known. If parental data are not available, an independent sequencing run with a different technology or library preparation can provide validation data.

For clinical applications, the validation should include the specific variants of interest. The rare disease study using nanopore sequencing established diagnostic variants in 11 probands, with diverse underlying genetic causes including de novo and compound heterozygous variants, large-scale structural variants, and epigenetic modifications [9]. The validation of these variants required careful comparison to the clinical presentation and inheritance pattern.

### Reproducibility and Documentation

Reproducibility requires documentation of all parameters and data versions. The nf-core documentation provides standards for community pipeline usage and configuration that emphasize reproducibility [5]. The Galaxy Training Network offers accessible workflow training that includes reproducibility as a core principle [4].

The Bioconductor project provides official documentation for reproducible genomic analysis workflows [3]. The Carpentries lessons provide foundational training in data organization and version control that supports reproducible research [6]. These resources can be adapted for assembly projects to ensure that the analysis can be reproduced by other laboratories.

## Professional Escalation Criteria

### When to Seek Additional Expertise

Several situations warrant consultation with a bioinformatics specialist or a core facility. If the assembly produces highly fragmented contigs despite adequate coverage, the problem may be in the library preparation or sequencing instead of the assembly parameters. If the phasing accuracy is below acceptable thresholds for the intended application, additional data types or a different assembly strategy may be needed.

For clinical applications, the escalation criteria should be defined before the analysis begins. If the assembly cannot resolve a variant of clinical interest, the case should be escalated to a molecular pathologist or clinical geneticist. The rare disease study demonstrated that long-read sequencing could enhance diagnostic yield, but the interpretation of variants required clinical expertise [9].

### Data Quality Red Flags

Certain patterns in the data indicate problems that require escalation. A sudden drop in read length or quality in the middle of a sequencing run may indicate a flow cell problem. An unexpected change in coverage across the genome may indicate sample contamination or library preparation issues. A high rate of chimeric reads may indicate problems with the library preparation or the sequencing platform.

The NCBI provides resources for sequence data quality assessment and troubleshooting [1]. The EMBL-EBI training resources provide guidance on data quality assessment and analysis best practices [2]. These resources can help laboratories identify and resolve data quality issues before they compromise the assembly.

## Frequently Asked Questions

### What is the difference between primary contigs and haplotype-resolved contigs?

Primary contigs represent one consensus sequence per genomic region, which may be a mosaic of the two haplotypes in regions where they differ. Haplotype-resolved contigs represent the two parental chromosomes as separate sequences, preserving the distinction between maternal and paternal alleles. The primary contig approach is simpler but loses haplotype information, while the haplotype-resolved approach preserves the phase of variants.

### How much sequencing coverage is needed for haplotype-resolved assembly?

The coverage requirement depends on the sequencing platform and assembler. For HiFi reads with hifiasm, 30 to 40-fold coverage is typically sufficient. For continuous long reads with HiCanu or Falcon-Unzip, 50 to 60-fold coverage is a common starting point. The Canu assembler halves depth-of-coverage requirements compared to its predecessor, but sufficient depth is still needed for reliable overlap detection [8].

### Can haplotype-resolved assembly be done without parental samples?

Yes, haplotype-resolved assembly can be done without parental samples using Hi-C data or by relying on the heterozygosity of the sample itself. FALCON-Phase uses Hi-C chromatin interaction data to extend phase blocks to chromosome scale without parental data [11]. The accuracy of this approach depends on the heterozygosity of the sample, with higher accuracy in more heterozygous species [11].

### What is the advantage of HiFi reads over continuous long reads for phasing?

HiFi reads provide higher accuracy, which simplifies the assembly graph and improves haplotype separation. The original HiFi demonstration showed 99.8% accuracy with an average read length of 13.5 kilobases [10]. This accuracy means that individual reads can be trusted for variant detection, which reduces the ambiguity in the assembly graph. Continuous long reads are longer but less accurate, requiring more complex error correction and consensus approaches.

### How does Hi-C phasing compare to trio phasing in accuracy?

Trio phasing provides definitive assignment of haplotypes because the inheritance pattern is known. Hi-C phasing is less definitive and its accuracy depends on the heterozygosity of the sample. In the Vertebrate Genomes Project validation, FALCON-Phase achieved 97% accuracy for cow and zebra finch compared to 80 to 91% for human [11]. Trio phasing should be preferred when parental samples are available and definitive haplotype assignment is required.

### What causes phase switch errors and how can they be detected?

Phase switch errors occur when the assignment of variants to haplotypes is correct within a phase block but the orientation of the block is flipped relative to the true haplotypes. These errors are invisible in most quality metrics because the assembly is internally consistent. Detection requires external validation data, such as parental genotypes or independent sequencing data.

### How should assembly quality be validated for clinical applications?

Clinical validation should include comparison to independently determined haplotypes, assessment of the specific variants of interest, and review by a clinical geneticist. The rare disease study using nanopore sequencing established diagnostic variants in 11 probands, with diverse underlying genetic causes [9]. The validation process should be defined before the analysis begins and should include escalation criteria for unresolved cases.

### What are the main limitations of current haplotype-resolved assembly methods?

The main limitations are the inability to resolve highly repetitive regions, the dependence of phasing accuracy on heterozygosity, and the requirement for sufficient coverage and data quality. Highly repetitive regions such as centromeres and acrocentric short arms remained intractable until recently [7]. The accuracy of Hi-C phasing varies by species, with lower accuracy in low-heterozygosity samples [11].

## Related Bioinformatics Guides

- [Long-Read Sequencing for De Novo Assembly of Complex Genomes: Case Studies and Best Practices](/knowledge/bioinformatics/long-read-sequencing-for-de-novo-assembly-of-complex-genomes-case-studies-and-best-practices)
- [De Novo Genome Assembly with Long Reads: A Practical Workflow](/knowledge/bioinformatics/de-novo-genome-assembly-with-long-reads-a-practical-workflow)
- [Long-Read Metagenome Assembly: Overcoming Challenges with Nanopore and PacBio Data](/knowledge/bioinformatics/long-read-metagenome-assembly-overcoming-challenges-with-nanopore-and-pacbio-data)
- [Evaluating Metagenomic Assembly Tools: A Benchmarking Framework for Short-Read and Long-Read Data](/knowledge/bioinformatics/evaluating-metagenomic-assembly-tools-a-benchmarking-framework-for-short-read-and-long-read-data)
- [Hybrid Genome Assembly: Combining Short and Long Reads for Better Results](/knowledge/bioinformatics/hybrid-genome-assembly-combining-short-and-long-reads-for-better-results)

## References and Further Reading

- [NCBI Data Resources](https://www.ncbi.nlm.nih.gov/). National Center for Biotechnology Information.
- [EMBL-EBI Training](https://www.ebi.ac.uk/training). European Bioinformatics Institute.
- [Bioconductor](https://bioconductor.org/). Bioconductor Project.
- [Galaxy Training Network](https://training.galaxyproject.org/). Galaxy Project.
- [nf-core Documentation](https://nf-co.re/docs). nf-core.
- [The Carpentries Lessons](https://carpentries.org/lessons). The Carpentries.
- [Long-Read DNA Sequencing: Recent Advances and Remaining Challenges.](https://pubmed.ncbi.nlm.nih.gov/37075062). Annual review of genomics and human genetics, 2023.
- [Canu: scalable and accurate long-read assembly via adaptive k-mer weighting and repeat separation.](https://pubmed.ncbi.nlm.nih.gov/28298431). Genome research, 2017.
- [Advancing long-read nanopore genome assembly and accurate variant calling for rare disease detection.](https://pubmed.ncbi.nlm.nih.gov/39862869). American journal of human genetics, 2025.
- [Accurate circular consensus long-read sequencing improves variant detection and assembly of a human genome.](https://pubmed.ncbi.nlm.nih.gov/31406327). Nature biotechnology, 2019.
- [Extended haplotype-phasing of long-read de novo genome assemblies using Hi-C.](https://pubmed.ncbi.nlm.nih.gov/33911078). Nature communications, 2021.

> This article is educational and does not replace validated analysis plans, institutional policy, clinical interpretation, or specialist review.