# Haplotype-Resolved Assembly with PacBio HiFi: A Practical Workflow Using HiCanu and Falcon-Unzip ## Scope and Reader Context Haplotype-resolved assembly from PacBio HiFi data addresses a specific problem: standard diploid genome assemblies collapse homologous chromosomes into a single consensus sequence, obscuring allele-specific variation, structural rearrangements, and heterozygosity patterns that matter for functional genomics and breeding. This article provides a practical workflow for researchers who have PacBio HiFi reads and need phased assemblies, comparing two established tools: HiCanu and Falcon-Unzip. The workflow covers raw read preparation, parameter selection, assembly execution, quality assessment, and interpretation of outputs. It assumes working familiarity with the Linux command line and basic genome assembly concepts. The guidance applies to diploid organisms with moderate heterozygosity, with explicit notes on limitations for polyploids and highly repetitive genomes. ## Context and Rationale for Haplotype-Resolved Assembly ### Why Standard Assemblies Are Insufficient Most reference genomes represent a single consensus haplotype. When two parental chromosomes differ substantially, a collapsed assembly merges divergent regions into chimeric sequences or drops one allele entirely. This creates systematic errors in downstream analyses including variant calling, expression quantification, and comparative genomics. High-quality reference genomes are essential to effectively characterize genomic drivers of speciation, phenotypic diversity, and disease causality, and collapsed assemblies undermine each of these applications. Haplotype-resolved assemblies preserve both parental copies as separate sequences. This matters for heterozygous organisms where allelic differences drive phenotype. In outcrossing forage crops and animal breeds, allele-specific expression and structural variation between haplotypes contribute to trait variation. A haplotype-resolved assembly enables allele-aware analyses that a collapsed assembly cannot support. ### PacBio HiFi Data Characteristics PacBio HiFi reads are circular consensus sequences generated from single molecules. Each read has accuracy above 99.9 percent with lengths typically ranging from 10 to 25 kilobases. This combination of long length and high accuracy makes HiFi reads suitable for both assembly and phasing. The error profile is largely random instead of systematic, which simplifies assembly algorithms compared to older long-read technologies with higher error rates. HiFi data alone can produce haplotype-resolved assemblies for many diploid genomes. Some genomes require additional data types. Larger complex genomes often require integration of long-read DNA sequencing with additional genomic data, such as chromosome conformation capture (Hi-C or CiFi) to generate phased chromosome-scale assemblies, though this requires multiple sequencing platforms or the construction of multiple long-read sequencing libraries. A single-library approach combining PacBio HiFi and CiFi sequencing can efficiently produce high-quality contiguous chromosome-scale diploid genome assemblies, as demonstrated in vole genomes where this strategy generated haplotype-resolved, chromosome-scale 2.3 Gbp genomes with high quality and completeness. ### When Haplotype Resolution Is Necessary Consider haplotype-resolved assembly when any of the following apply: - The organism is highly heterozygous, including outcrossing plants, many animal species, and F1 crosses - Allele-specific expression analysis is planned - Structural variant discovery between haplotypes is a research goal - A pangenome representation is intended - The species has no existing reference or the existing reference is from a different individual For inbred lines or homozygous samples, haplotype resolution adds computational cost without proportional benefit. Assess heterozygosity from k-mer analysis of raw reads before committing to a phased assembly strategy. ## Core Principles of HiFi Assembly and Phasing ### Read Overlap and Contig Construction HiFi assemblers build contigs by finding overlaps between reads. The high accuracy of HiFi reads permits sensitive overlap detection because sequencing errors do not obscure true sequence identity. Both HiCanu and Falcon-Unzip use overlap-based approaches, though they differ in implementation details. HiCanu is the HiFi-optimized version of the Canu assembler. It applies error correction tailored to HiFi error profiles, then constructs contigs from corrected reads. For heterozygous genomes, HiCanu can produce primary and alternate assemblies, where the primary assembly represents one haplotype and the alternate contains haplotype-specific contigs that did not collapse into the primary. Falcon-Unzip takes a different path. Falcon first produces a primary assembly, then Falcon-Unzip phases the primary contigs into haplotypes using read alignment and variant detection. The output includes primary contigs and haplotype-specific contigs (hap1 and hap2) that represent the two parental copies. ### Phasing Strategies Phasing assigns sequence variants to parental chromosomes. Two broad strategies exist: Reference-based phasing aligns reads to a reference and uses variant linkage to assign haplotypes. This approach suffers from reference bias when the sample diverges substantially from the reference. Graph-based phasing builds a variation graph from the reads themselves, avoiding reference bias but potentially producing imbalanced haplomes when structural variation between haplotypes is abundant. Current methods to generate haplotype-resolved assemblies for highly heterozygous genomes suffer from reference bias when relying on reference-based phasing or may generate imbalanced haplomes when relying on graph-based phasing. Newer tools such as PhaseGrass combine reference-based phasing with haplotype-specific k-mers to partition long reads or unitigs to corresponding haplotypes, avoiding both reference bias and uneven haplotype partitioning. This approach generated chromosome-level, haplotype-resolved assemblies for highly heterozygous grass species and binned more reads to haplotypes than standard phasing tools while producing balanced haplomes compared to graph-based assemblers that produced largely imbalanced haplomes due to abundant structural variations between haplotypes. HiCanu and Falcon-Unzip represent earlier approaches to phasing. Both can produce useful haplotype-resolved assemblies, but newer tools may outperform them for highly heterozygous or structurally complex genomes. The choice depends on the specific genome characteristics and the available computational resources. ### Quality Metrics for Phased Assemblies Several metrics assess haplotype-resolved assembly quality: - Contig N50: the length at which half the assembly is in contigs of that length or longer - BUSCO completeness: the fraction of conserved single-copy orthologs present in the assembly - QV (consensus accuracy): the estimated base accuracy of the assembly - Haplotype balance: the ratio of sequence content between the two haplotypes - Switch errors: positions where the phasing incorrectly switches between parental chromosomes Technical validation of haplotype-resolved assemblies typically includes BUSCO, LTR Assembly Index, Inspector, and Merqury analyses, all supporting the accuracy and completeness of the assembly. Merqury uses k-mers from raw reads to evaluate consensus accuracy and completeness without requiring a reference genome. ## At a Glance: Workflow Decision Table | Decision Point | HiCanu | Falcon-Unzip | Notes | |---|---|---|---| | Input requirement | HiFi reads in FASTQ or BAM format | HiFi reads in FASTQ format, pre-assembly with Falcon | Both accept standard PacBio output | | Best suited for | Moderate heterozygosity, genomes up to several Gbp | Moderate heterozygosity, genomes with clear haplotype blocks | Highly heterozygous genomes may need newer tools | | Output structure | Primary plus alternate contigs | Primary plus hap1 and hap2 contigs | Both require post-assembly QC | | Computational demand | High memory, parallelizable | High memory for Falcon step, additional compute for Unzip | Plan for 1 TB RAM for large genomes | | Phasing approach | Overlap-based with haplotype-aware contig construction | Read alignment and variant-based phasing of primary contigs | Falcon-Unzip depends on read depth for phasing accuracy | | Recommended QC | Merqury, BUSCO, QUAST | Merqury, BUSCO, QUAST | Same QC pipeline works for both | | Typical runtime | Days for mammalian genomes | Days to weeks depending on depth | Runtime scales with genome size and coverage | ## Practical Workflow: From Raw Reads to Phased Assembly ### Step 1: Data Preparation and Quality Assessment Start with raw PacBio HiFi reads in FASTQ format. Most sequencing facilities deliver demultiplexed, trimmed reads. Verify read quality before assembly. Run these checks: 1. Count reads and total bases 2. Calculate read N50 3. Assess read length distribution 4. Check for adapter contamination 5. Estimate genome size and heterozygosity using k-mer analysis Tools for k-mer analysis include Jellyfish for k-mer counting and GenomeScope for genome size and heterozygosity estimation. These estimates inform coverage calculations and assembly strategy. Coverage calculation: divide total bases by estimated genome size. For HiFi assembly, 30x to 50x coverage is typical for diploid genomes. Higher coverage improves phasing but increases computational cost. Coverage below 25x may produce fragmented assemblies or incomplete phasing. ### Step 2: Read Filtering and Trimming PacBio HiFi reads generally arrive clean, but filtering improves assembly quality: - Remove reads shorter than 1 kb, as very short reads add little assembly information - Remove reads with average quality below Q20 - Trim adapter sequences if present - Remove reads with excessive homopolymer errors, which can indicate sequencing artifacts HiCanu includes internal read correction and trimming steps. Falcon-Unzip also performs read correction. Pre-filtering reduces the burden on these steps and can improve runtime. ### Step 3: Assembly with HiCanu HiCanu is invoked with the canu command using the pacbio-hifi option. The basic command structure: ``` canu -p output_prefix -d output_directory genomeSize=1e9 -pacbio-hifi reads.fastq ``` Key parameters: - genomeSize: estimated genome size, used for coverage calculations and read correction - minReadLength: minimum read length to include, default 1000 - minOverlapLength: minimum overlap for read correction, default 500 for HiFi - corOutCoverage: coverage to use for correction, default 40 - correctedErrorRate: expected error rate after correction, default 0.045 for HiFi For heterozygous genomes, HiCanu automatically produces a primary assembly and an alternate assembly. The primary assembly contains the best-supported haplotype. The alternate contains haplotype-specific contigs. Parameter adjustments for specific cases: - High heterozygosity: increase corOutCoverage to 60 or higher to ensure both haplotypes are represented in correction - Repetitive genomes: increase minOverlapLength to reduce false overlaps in repeats - Low coverage: reduce corOutCoverage to avoid over-correction ### Step 4: Assembly with Falcon-Unzip Falcon-Unzip requires two stages. First, Falcon produces a primary assembly. Second, Falcon-Unzip phases the primary contigs. Falcon configuration uses a configuration file specifying parameters: ``` [General] input_fofn = input.fofn input_type = raw pa_DBdir = preads genome_size = 1000000000 seed_coverage = 30 length_cutoff = 12000 ``` The input.fofn file lists the paths to read files. Falcon aligns reads, corrects errors, and constructs a primary assembly. Falcon-Unzip then aligns reads back to the primary contigs, detects variants, and phases contigs into haplotypes. The output includes: - Primary contigs - Haplotype-specific contigs (hap1 and hap2) - Phasing information for variants Falcon-Unzip requires sufficient read depth for accurate phasing. Coverage below 40x may produce incomplete phasing. The phasing quality depends on the density of heterozygous variants along the genome. ### Step 5: Quality Assessment Run the same quality assessment for both assembler outputs: 1. BUSCO completeness: assess gene content completeness using the appropriate lineage dataset 2. Merqury: evaluate consensus accuracy and completeness using k-mers from raw reads 3. QUAST: compute contig statistics including N50, total length, and GC content 4. Inspector: assess assembly errors including misjoins and base errors 5. LTR Assembly Index: evaluate assembly quality in repetitive regions For haplotype-resolved assemblies, also assess: - Haplotype balance: compare total sequence length between haplotypes - Switch error rate: estimate using read phasing information - Allele representation: verify that heterozygous variants are present in both haplotypes BUSCO completeness above 95 percent is typical for high-quality assemblies. QV above 50 indicates high consensus accuracy. Haplotype balance should be near 1:1 for diploid organisms, though some imbalance is expected in regions with structural variation between haplotypes. ### Step 6: Scaffolding and Chromosome-Scale Assembly Contig-level assemblies from HiCanu or Falcon-Unzip may be sufficient for many analyses. Chromosome-scale assembly requires additional data: - Hi-C data for chromatin conformation scaffolding - Genetic maps for linkage-based anchoring - Optical maps for long-range ordering Hi-C scaffolding assigns contigs to chromosomes and orders them within chromosomes. Tools such as HiRise, SALSA, and 3D-DNA perform this scaffolding. The resulting chromosome-scale assembly enables synteny analysis and comparative genomics. A single-library approach combining PacBio HiFi and CiFi sequencing can produce chromosome-scale diploid assemblies without separate Hi-C library construction. This strategy generated haplotype-resolved, chromosome-scale assemblies for vole genomes with high quality and completeness, simplifying the workflow and reducing cost. For organisms with existing genetic maps, anchoring contigs to the map can validate scaffolding and resolve ordering ambiguities. Synteny analyses and genetic map anchoring further confirm structural integrity and chromosome phasing. ### Step 7: Annotation and Downstream Analysis After assembly and scaffolding, annotation identifies genes, repeats, and other functional elements. Structural annotation predicts gene models using evidence from transcript alignments and ab initio prediction. Functional annotation assigns putative functions to predicted genes. For haplotype-resolved assemblies, annotation should be performed on both haplotypes separately. This enables allele-aware analyses including: - Allele-specific expression - Haplotype-specific structural variants - Comparative analysis of gene content between haplotypes A haplotype-resolved, chromosome-scale assembly of tetraploid sainfoin resolved all 28 pseudochromosomes corresponding to 4 haplotypes of the 7 base chromosomes, with high contiguity and gene completeness. The assembly supported annotation of over 117,000 high-confidence protein-coding genes and enabled allele-specific expression analysis showing largely balanced expression across haplotypes. ## Options and Tradeoffs: HiCanu versus Falcon-Unzip ### Assembly Quality Comparison Both HiCanu and Falcon-Unzip produce high-quality haplotype-resolved assemblies for many genomes. The choice depends on genome characteristics and available resources. HiCanu advantages: - Integrated read correction and assembly in a single pipeline - Automatic primary and alternate assembly generation - Well-documented parameter space - Active maintenance and community support HiCanu limitations: - Higher memory requirements than some alternatives - Runtime scales with genome size and coverage - Alternate assembly may be incomplete for highly heterozygous genomes Falcon-Unzip advantages: - Clear separation of assembly and phasing stages - Explicit variant-based phasing that can be inspected - Useful for genomes with clear haplotype blocks Falcon-Unzip limitations: - Requires two-stage execution with separate configuration - Phasing quality depends on read depth and variant density - Less actively maintained than some alternatives ### Computational Resource Considerations Both assemblers require substantial computational resources for large genomes. Plan for: - 500 GB to 1 TB RAM for mammalian-sized genomes - 16 to 64 CPU cores for parallel execution - 1 to 5 TB temporary disk space - Days to weeks of wall-clock time depending on genome size and coverage Cloud computing or institutional clusters are appropriate for large assemblies. Local workstations may suffice for small genomes below 500 Mbp. ### Newer Alternatives Hifiasm is a newer assembler that produces haplotype-resolved assemblies directly from HiFi reads. It often outperforms HiCanu and Falcon-Unzip in contiguity and phasing quality. However, Hifiasm can produce imbalanced haplomes for genomes with abundant structural variation between haplotypes. PhaseGrass addresses this limitation by combining reference-based phasing with haplotype-specific k-mers. It generated balanced haplomes for highly heterozygous grass species where Hifiasm produced largely imbalanced haplomes. PhaseGrass also binned more reads to haplotypes than standard phasing tools. For researchers starting new assemblies, consider Hifiasm or PhaseGrass as primary options. HiCanu and Falcon-Unzip remain useful for validation, comparison, and specific use cases where their output structure is preferred. ## Observations and Measurements ### Recording Assembly Parameters Document all assembly parameters and decisions. This record enables reproducibility and troubleshooting. Include: - Software versions for all tools - Parameter values and rationale for changes from defaults - Read statistics including total bases, read N50, and coverage - Genome size and heterozygosity estimates - Computational resources used and runtime - Quality metrics for the final assembly Version control for assembly scripts and configuration files ensures reproducibility. Containerized workflows using Docker or Singularity can capture the software environment. ### Measuring Assembly Quality Track these measurements throughout the assembly process: | Metric | Target | Measurement Tool | |---|---|---| | Read N50 | 10 to 25 kb for HiFi | seqkit, fastqc | | Coverage | 30x to 50x | Calculated from read bases and genome size | | Contig N50 | Species-dependent, higher is better | QUAST | | BUSCO completeness | Above 95 percent | BUSCO | | QV | Above 50 | Merqury | | LTR Assembly Index | Above 20 for reference quality | LTR_retriever | | Haplotype balance | Near 1:1 for diploids | Assembly statistics | Record these metrics at each assembly stage. Compare metrics between HiCanu and Falcon-Unzip outputs to identify the better assembly for downstream use. ### Interpreting Phasing Quality Phasing quality affects downstream analyses. Low phasing quality produces switch errors where haplotypes are incorrectly assigned. This creates false heterozygous variants and incorrect haplotype sequences. Assess phasing quality by: 1. Counting heterozygous variants per 100 kb window 2. Checking for regions with unusually low variant density, which may indicate collapsed haplotypes 3. Comparing haplotype sequences for regions of unexpected similarity 4. Validating phasing against known pedigree data if available Regions with low variant density may represent collapsed paralogs or recent segmental duplications. These regions require additional validation. ## Records and Documentation ### Assembly Project Records Maintain a structured record for each assembly project: 1. Sample metadata: species, individual identifier, tissue source, DNA extraction method 2. Sequencing metadata: library preparation protocol, sequencing platform, run dates, yield 3. Read processing: filtering parameters, quality thresholds, final read counts 4. Assembly parameters: tool versions, parameter values, rationale for changes 5. Quality assessment: all metrics with tool versions and dates 6. Assembly versions: file names, locations, and checksums 7. Downstream analyses: annotation, scaffolding, and validation results This record supports publication requirements and enables future updates when improved tools become available. ### Reproducibility Practices Reproducible assembly workflows use: - Version-controlled scripts and configuration files - Containerized software environments - Documented parameter choices with justification - Raw data preservation with checksums - Analysis logs capturing tool output and warnings Training resources from [The Carpentries](https://carpentries.org/lessons) cover foundational computing, data, shell, Git, and programming skills that support reproducible bioinformatics workflows. The [Galaxy Training Network](https://training.galaxyproject.org/) provides accessible workflow training and analysis tutorials for reproducibility context. [Bioconductor](https://bioconductor.org/) offers official package, workflow, installation, and reproducible genomic-analysis documentation. [EMBL-EBI Training](https://www.ebi.ac.uk/training) provides bioinformatics learning pathways and data-resource training for practical analysis education. ## Common Failure Patterns and Troubleshooting ### Fragmented Assemblies Symptoms: low contig N50, many small contigs, incomplete BUSCO Possible causes: - Insufficient coverage below 25x - High heterozygosity causing haplotype separation into many small contigs - Repetitive regions causing assembly breaks - Contaminated reads from other organisms Remedies: - Increase coverage with additional sequencing - Adjust heterozygosity parameters in the assembler - Filter contaminant reads using taxonomic classification - Use a different assembler better suited to the genome characteristics ### Imbalanced Haplotypes Symptoms: one haplotype much larger than the other, many variants missing from one haplotype Possible causes: - Structural variation between haplotypes causing one haplotype to assemble poorly - Insufficient read depth for phasing - Reference bias in phasing approaches Remedies: - Increase coverage to improve representation of both haplotypes - Use a phasing approach that avoids reference bias - Validate with k-mer analysis to check for missing haplotype content ### High Switch Error Rate Symptoms: phasing inconsistencies, false heterozygous variants, haplotype sequences that mix parental alleles Possible causes: - Low variant density in some regions - Insufficient read depth for phasing - Structural variation confusing phasing algorithms Remedies: - Increase coverage - Use long-range phasing data such as Hi-C or linked reads - Apply phasing polishing tools to correct switch errors ### Contamination Symptoms: unexpected GC content, foreign BUSCO hits, assembly size exceeding expectations Possible causes: - Sample contamination during DNA extraction - Index hopping during multiplexed sequencing - Laboratory cross-contamination Remedies: - Run taxonomic classification on reads before assembly - Check mitochondrial or chloroplast reads for expected origin - Repeat DNA extraction and sequencing if contamination is confirmed ### Memory or Runtime Failures Symptoms: assembler crashes, out-of-memory errors, excessive runtime Possible causes: - Genome size underestimated - Coverage higher than expected - Insufficient computational resources Remedies: - Reduce coverage using read subsampling - Increase memory allocation - Use distributed computing or cloud resources - Adjust parameters to reduce computational demand ## Limitations and Interpretation Boundaries ### What Haplotype-Resolved Assemblies Cannot Provide Haplotype-resolved assemblies from HiFi data have specific limitations: - They do not resolve all repetitive regions, particularly centromeres and large satellite arrays - They may not fully phase all variants, especially in regions with low heterozygosity - They represent the sequenced individual, not the species consensus - They may contain assembly errors in complex regions that require manual curation Pangenome projects have enhanced understanding of human genomic and genetic diversity, but repetitive regions remain challenging to assemble and are medically important. Recent work generated 20 near-complete haplotypes from 10 individuals using three complementary long-read and long-range datasets, substantially improving the average reconstruction rate of complete haplotypes within segmentally duplicated complex regions. This demonstrates that complex regions require additional data beyond standard HiFi assembly. ### Polyploid Genomes The workflow described here applies to diploid genomes. Polyploid genomes require different approaches. Tetraploid sainfoin assembly resolved all 28 pseudochromosomes corresponding to 4 haplotypes of the 7 base chromosomes using PacBio HiFi, Oxford Nanopore, Illumina short read, and Hi-C data. This multi-platform approach was necessary for the tetraploid genome. For polyploid species, consult specialized tools and workflows designed for polyploid assembly. Standard diploid assemblers may collapse or misassign haplotypes in polyploid genomes. ### Structural Variant Detection Haplotype-resolved assemblies enable structural variant detection between haplotypes. Comparing the two new vole genomes identified complex structural changes impacting a gene previously implicated in pair bonding, including a species-specific duplication missing from the existing reference genome. These divergent genomic features offer new avenues of investigation related to behavioral divergence. Structural variant detection from assemblies requires careful validation. Assembly errors can create false structural variants. Validate candidate variants using read alignments and PCR where possible. ### Reference Bias Reference-based phasing approaches introduce reference bias when the sample diverges from the reference. This bias can cause alleles matching the reference to be overrepresented and divergent alleles to be missed. Graph-based and k-mer-based phasing approaches reduce this bias. For species without a close reference, use phasing approaches that do not depend on a reference genome. HiCanu and Falcon-Unzip both perform reference-free phasing, though Falcon-Unzip uses the primary assembly as an internal reference. ## Quality Controls and Validation ### K-mer-Based Validation Merqury uses k-mers from raw reads to evaluate assembly quality. It provides: - Consensus accuracy (QV) - Completeness - Error rates - Haplotype-specific metrics for phased assemblies Run Merqury on both HiCanu and Falcon-Unzip outputs. Compare QV and completeness to identify the better assembly. ### Gene Content Validation BUSCO assesses gene content completeness using conserved single-copy orthologs. Select the appropriate lineage dataset for the organism. Run BUSCO on both haplotypes separately and on the combined assembly. BUSCO completeness above 95 percent indicates good gene representation. Lower completeness may indicate assembly gaps in gene-rich regions or haplotype collapse. ### Read Alignment Validation Align raw reads back to the assembly to check for: - Reads spanning assembly breaks - Reads with unexpected alignment patterns - Coverage uniformity across the assembly Unexpected coverage patterns may indicate collapsed repeats or misassembled regions. Inspector provides automated assembly error assessment. ### LTR Assembly Index The LTR Assembly Index (LAI) assesses assembly quality in repetitive regions. Reference-grade assemblies typically have LAI above 20. Lower LAI indicates incomplete assembly of long terminal repeat retrotransposons. For haplotype-resolved assemblies, calculate LAI for each haplotype separately. Differences in LAI between haplotypes may indicate assembly bias. ## Safety and Regulatory Context ### Data Management Genome assembly projects generate large data volumes. Follow institutional data management policies for storage, backup, and sharing. Raw sequencing data may be subject to data sharing agreements or consent restrictions. Public databases such as [NCBI](https://www.ncbi.nlm.nih.gov/) provide official descriptions of databases, search systems, sequence resources, and analysis services. Deposit assemblies and raw data in appropriate public repositories to support reproducibility and community access. ### Computational Ethics Large assembly projects consume substantial computational resources. Use resources efficiently: - Right-size computational requests to avoid waste - Share intermediate data where appropriate - Document resource usage for future planning ### Export and Transfer Considerations Some genome data may be subject to export controls or material transfer agreements. Check institutional policies before transferring data across borders or to collaborators. ## Professional Escalation Criteria ### When to Seek Expert Assistance Escalate to a bioinformatics specialist or assembly expert when: - Assembly fails repeatedly with standard parameters - Quality metrics remain below acceptable thresholds after parameter adjustment - The genome has unusual characteristics such as extreme heterozygosity or polyploidy - Structural variants are the primary research focus and require manual curation - The assembly will serve as a community reference genome ### When to Consider Additional Data Add more data when: - Coverage is below 25x and additional sequencing is feasible - Phasing quality is poor and long-range data such as Hi-C could improve it - Repetitive regions are unresolved and additional platforms could help - The genome is highly heterozygous and current tools produce imbalanced haplotypes ### When to Change Tools Switch to a different assembler when: - Current assembler produces fragmented assemblies despite parameter adjustment - Haplotype imbalance persists with current tools - Newer tools offer demonstrated improvements for similar genomes - Current tool is no longer maintained and has known issues ## Decision Framework for Assembler Selection and Resource Allocation ### Structured Triage Before Assembly Begins The choice between HiCanu and Falcon-Unzip should not be made after sequencing completes. A structured decision framework applied to raw read characteristics and project goals prevents wasted compute cycles and produces assemblies matched to downstream analysis requirements. This framework uses measurable read properties and explicit project constraints to select the assembler, allocate resources, and define success criteria before the first assembly command runs. ### Step 1: Characterize the Input Data Run a standardized read characterization protocol before any assembly decision. This protocol produces the numeric inputs needed for the decision framework. 1. Count total reads and total bases using seqkit stats or an equivalent tool 2. Calculate read N50 and read length distribution 3. Estimate genome size and heterozygosity using k-mer analysis with Jellyfish and GenomeScope 4. Compute coverage by dividing total bases by the estimated genome size 5. Assess read quality distribution and flag any reads below Q20 6. Check for adapter contamination using fastqc or a dedicated adapter detection tool Record these values in a structured format. The decision framework uses coverage, heterozygosity, and estimated genome size as primary inputs. Read N50 and quality distribution inform secondary decisions about filtering parameters. ### Step 2: Apply the Assembler Selection Matrix Use the following decision matrix to select the assembler based on measured read characteristics and project constraints. The matrix assumes diploid genomes and moderate repeat content. Polyploid genomes require specialized tools outside this framework. | Read Characteristic | HiCanu Recommended | Falcon-Unzip Recommended | Either Acceptable | |---|---|---|---| | Coverage 25x to 35x | Yes, with corOutCoverage adjusted to 30 | Yes, but phasing may be incomplete | Yes | | Coverage 35x to 50x | Yes, default parameters | Yes, default parameters | Yes | | Coverage above 50x | Yes, reduce corOutCoverage to 40 | Yes, subsample reads to 50x | Yes | | Heterozygosity below 0.5 percent | Yes | Yes | Yes | | Heterozygosity 0.5 to 1.5 percent | Yes, increase corOutCoverage | Yes, verify variant density | Yes | | Heterozygosity above 1.5 percent | Caution, may produce fragmented alternate | Caution, may produce imbalanced haplotypes | No, consider Hifiasm or PhaseGrass | | Genome size below 500 Mbp | Yes, low memory footprint | Yes | Yes | | Genome size 500 Mbp to 3 Gbp | Yes, plan 500 GB to 1 TB RAM | Yes, plan 500 GB to 1 TB RAM | Yes | | Genome size above 3 Gbp | Yes, plan 1 TB RAM or more | Caution, Falcon stage may be slow | Yes with cluster resources | | Structural variant analysis primary goal | Yes, alternate assembly useful | Yes, hap1 and hap2 contigs useful | Yes | | Allele-specific expression planned | Yes | Yes | Yes | | Community reference genome intended | Yes | Yes | Yes, but consider newer tools | | Rapid turnaround required | Yes, single pipeline | No, two-stage execution | HiCanu preferred | Apply the matrix as a filter, not a prescription. If both assemblers are acceptable, run both when resources permit. If the matrix indicates caution for both tools, escalate to newer assemblers before committing compute resources. ### Step 3: Define Success Criteria Before Assembly Write explicit success criteria before launching the assembly. These criteria become the acceptance test for the final assembly and prevent indefinite parameter tweaking. Define minimum thresholds for: - Contig N50, based on published assemblies of related species or the estimated genome size - BUSCO completeness, typically above 95 percent for the appropriate lineage dataset - QV, typically above 50 for HiFi-based assemblies - Haplotype balance, typically between 0.8 and 1.2 for the ratio of hap1 to hap2 sequence content - Switch error rate, if estimable from available data Record these thresholds in the project record. If the assembly fails to meet thresholds after two parameter adjustment cycles, escalate according to the professional escalation criteria in the main workflow. ### Step 4: Allocate Computational Resources by Genome Size Resource allocation follows genome size and coverage. Use these planning figures as starting points, then adjust based on available infrastructure. For genomes below 500 Mbp: - 64 GB to 128 GB RAM - 16 to 32 CPU cores - 500 GB temporary disk space - 12 to 48 hours wall-clock time for HiCanu - 24 to 72 hours for Falcon-Unzip including both stages For genomes from 500 Mbp to 3 Gbp: - 256 GB to 512 GB RAM - 32 to 64 CPU cores - 2 to 5 TB temporary disk space - 2 to 7 days for HiCanu - 3 to 10 days for Falcon-Unzip For genomes above 3 Gbp: - 1 TB RAM or more - 64 or more CPU cores - 5 to 10 TB temporary disk space - 1 to 3 weeks for HiCanu - 2 to 4 weeks for Falcon-Unzip These figures assume 30x to 50x coverage. Higher coverage increases both memory and runtime requirements. Lower coverage reduces requirements but risks incomplete phasing. ### Step 5: Implement a Staged Assembly Protocol Run assemblies in stages with quality gates between stages. This protocol catches failures early and avoids wasting compute on doomed assemblies. Stage 1: Read filtering and subsampling - Filter reads below 1 kb and below Q20 - If coverage exceeds 60x, subsample to 50x using a random subsampling tool - Record the filtered read count and total bases Stage 2: Small-scale test assembly - Assemble a random subset of 5x to 10x coverage - Run BUSCO and QUAST on the test assembly - Verify that the test assembly produces expected contiguity and completeness - If the test assembly fails, adjust parameters before the full assembly Stage 3: Full assembly with the selected assembler - Launch the full assembly with parameters informed by the test assembly - Monitor runtime and resource usage - Record all parameter values and tool versions Stage 4: Quality assessment and acceptance testing - Run the full quality assessment suite from the main workflow - Compare metrics against the predefined success criteria - Document results in the project record Stage 5: Comparison assembly with the alternative tool - If resources permit, run the alternative assembler - Compare quality metrics between the two assemblies - Select the better assembly for downstream use This staged protocol adds a small test assembly step that many researchers skip. The test assembly costs a few hours and prevents multi-day failures from incorrect parameter choices. ### Step 6: Record Decisions and Outcomes Maintain a decision log that captures the rationale for each choice. This log supports troubleshooting and provides material for methods sections in publications. Record for each decision: - The decision made - The data that informed the decision - The expected outcome - The actual outcome - Any deviations from the expected outcome Example decision log entries: - Decision: Selected HiCanu over Falcon-Unzip - Data: Coverage 42x, heterozygosity 0.8 percent, genome size 1.2 Gbp - Expected outcome: Primary and alternate assemblies with balanced haplotypes - Actual outcome: Primary assembly N50 of 18 Mbp, alternate assembly 92 percent of primary length - Deviation: Alternate assembly shorter than expected, likely due to structural variation This log becomes part of the assembly project record described in the main workflow. ### Common Decision Framework Failures Several recurring mistakes undermine the decision framework. Recognize these patterns and correct them early. Failure pattern 1: Skipping the test assembly Researchers often skip the small-scale test assembly to save time. This decision frequently leads to multi-day full assemblies that fail from parameter errors that a test assembly would have revealed within hours. Always run the test assembly for genomes above 500 Mbp. Failure pattern 2: Ignoring heterozygosity estimates Some researchers proceed with default parameters without measuring heterozygosity. Highly heterozygous genomes assembled with default parameters produce fragmented assemblies or imbalanced haplotypes. Always run k-mer analysis before assembly and adjust parameters based on the heterozygosity estimate. Failure pattern 3: Overcommitting resources to the wrong assembler Falcon-Unzip requires two-stage execution and may need more wall-clock time than HiCanu for the same genome. Researchers with tight deadlines sometimes choose Falcon-Unzip without accounting for the additional time. Use the decision matrix to match assembler choice to timeline constraints. Failure pattern 4: Failing to define success criteria Without predefined success criteria, researchers accept assemblies that fail quality thresholds because they lack a reference point. Define success criteria before assembly and apply them consistently. Failure pattern 5: Not documenting parameter rationale Undocumented parameter changes create reproducibility problems and make troubleshooting difficult. Record the rationale for every parameter change from defaults. ### Integration with Existing Workflow Steps This decision framework operates before the assembly steps described in the main workflow. It produces the parameter choices and resource allocations used in Steps 3 and 4 of the practical workflow. The quality assessment in Step 5 of the main workflow provides the metrics that determine whether the decision framework produced the expected outcome. The framework also connects to the troubleshooting section. When assemblies fail, return to the decision log to identify which input assumptions were incorrect. For example, if the heterozygosity estimate was too low, the assembly may show signs of haplotype collapse. Correct the estimate and adjust parameters accordingly. ### Validation of the Decision Framework The decision framework aligns with published assembly projects that demonstrate the importance of matching assembler choice to genome characteristics. Highly heterozygous grass genomes required specialized phasing approaches because standard graph-based phasing produced imbalanced haplomes due to abundant structural variation between haplotypes. The decision matrix flags this situation by recommending caution for both HiCanu and Falcon-Unzip when heterozygosity exceeds 1.5 percent. Similarly, complex genomes with segmental duplications required multiple complementary long-read and long-range datasets to achieve complete haplotype reconstruction. The decision framework accounts for this by recommending additional data types when repetitive regions are a known challenge. For routine diploid genomes with moderate heterozygosity, both HiCanu and Falcon-Unzip produce useful assemblies. The decision framework helps researchers select between them based on project-specific constraints including timeline, computational resources, and downstream analysis requirements. ### When to Bypass the Framework The decision framework assumes standard diploid genomes with moderate repeat content. Bypass the framework and escalate to specialized approaches when: - The genome is polyploid, requiring specialized polyploid assembly tools - The genome has extreme heterozygosity above 2 percent - The genome has massive repeat content that defeats standard assemblers - The project requires chromosome-scale assembly and Hi-C or CiFi data are available - The assembly will serve as a community reference requiring manual curation In these cases, consult the professional escalation criteria in the main workflow and consider newer tools such as Hifiasm or PhaseGrass, or multi-platform approaches that combine HiFi with Hi-C or CiFi data. ### Practical Implementation Checklist Use this checklist when implementing the decision framework: 1. Run read characterization and record all metrics 2. Apply the assembler selection matrix 3. Define success criteria with numeric thresholds 4. Allocate computational resources based on genome size 5. Run the staged assembly protocol with quality gates 6. Maintain a decision log with rationale for each choice 7. Compare outcomes against predefined success criteria 8. Document results for reproducibility This checklist ensures that the decision framework is applied consistently across projects and that the rationale for each assembly choice is preserved for troubleshooting and publication. ## Frequently Asked Questions ### What coverage of PacBio HiFi reads is needed for haplotype-resolved assembly? Coverage of 30x to 50x is typical for diploid genomes. Coverage below 25x may produce fragmented assemblies or incomplete phasing. Higher coverage improves phasing accuracy but increases computational cost. Calculate coverage by dividing total read bases by the estimated genome size. For highly heterozygous genomes, higher coverage within this range is recommended to ensure both haplotypes are well represented. ### How do I estimate genome size and heterozygosity before assembly? Use k-mer analysis on raw reads. Count k-mers with Jellyfish and analyze the distribution with GenomeScope. The k-mer distribution provides estimates of genome size, heterozygosity, and repeat content. These estimates inform coverage calculations and assembly strategy. For organisms with existing references, compare read coverage to the reference genome as a sanity check. ### Can I use the same workflow for Oxford Nanopore data? The workflow is optimized for PacBio HiFi data. Oxford Nanopore reads have different error profiles and length distributions. HiCanu supports Oxford Nanopore data with different parameter settings. Falcon-Unzip is designed for PacBio data. For Nanopore data, consider assemblers such as Flye or Shasta, and phasing tools such as Whatshap. The quality assessment steps apply to any assembly. ### How do I choose between HiCanu and Falcon-Unzip? Consider genome characteristics and available resources. HiCanu provides an integrated pipeline with automatic primary and alternate assembly generation. Falcon-Unzip offers explicit variant-based phasing that can be inspected. For highly heterozygous genomes with abundant structural variation, newer tools such as Hifiasm or PhaseGrass may outperform both. Run both assemblers if resources permit and compare quality metrics to select the better assembly. ### What is the difference between primary and alternate assemblies? The primary assembly contains the best-supported haplotype for each genomic region. The alternate assembly contains haplotype-specific contigs that did not collapse into the primary. Alternate contigs may represent the second haplotype in regions where haplotypes diverged sufficiently to assemble separately. For downstream analyses, use the primary assembly for most applications and incorporate alternate contigs for allele-aware analyses. ### How do I validate that phasing is correct? Assess phasing quality using multiple approaches. Check heterozygous variant density across the genome, with low density regions indicating potential phasing problems. Compare haplotype sequences for unexpected similarity. Use Merqury to evaluate haplotype-specific k-mer completeness. If pedigree data are available, validate phasing against expected inheritance patterns. Long-range data such as Hi-C can confirm phasing at chromosome scale. ### What should I do if my assembly has imbalanced haplotypes? Imbalanced haplotypes indicate that one haplotype assembled better than the other. Increase coverage to improve representation of both haplotypes. Check for structural variation between haplotypes that may cause assembly bias. Consider using a phasing approach that avoids reference bias. Validate with k-mer analysis to check for missing haplotype content. If imbalance persists, consider newer tools designed for highly heterozygous genomes. ### How do I scaffold my haplotype-resolved assembly to chromosome scale? Use Hi-C data for chromatin conformation scaffolding. Tools such as HiRise, SALSA, and 3D-DNA assign contigs to chromosomes and order them within chromosomes. A single-library approach combining PacBio HiFi and CiFi sequencing can produce chromosome-scale assemblies without separate Hi-C library construction. Genetic maps can validate scaffolding and resolve ordering ambiguities. Synteny analysis against related species can confirm chromosome assignments. ## Related Bioinformatics Guides - [Metabolomics Data Analysis in R: A Practical Workflow](/knowledge/bioinformatics/metabolomics-data-analysis-in-r-a-practical-workflow) - [De Novo Genome Assembly with Long Reads: A Practical Workflow](/knowledge/bioinformatics/de-novo-genome-assembly-with-long-reads-a-practical-workflow) - [Proteomics Data Analysis in R: A Practical Workflow for Differential Expression and Visualization](/knowledge/bioinformatics/proteomics-data-analysis-in-r-a-practical-workflow-for-differential-expression-and-visualization) - [Long-Read Metagenome Assembly: Overcoming Challenges with Nanopore and PacBio Data](/knowledge/bioinformatics/long-read-metagenome-assembly-overcoming-challenges-with-nanopore-and-pacbio-data) - [Spatial Transcriptomics Data Analysis: A Practical Workflow from Raw Data to Biological Insights](/knowledge/bioinformatics/spatial-transcriptomics-data-analysis-a-practical-workflow-from-raw-data-to-biological-insights) ## 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. - [Single-library chromosome-scale diploid assemblies of vole genomes resolve a species-specific duplication implicated in pair bonding.](https://pubmed.ncbi.nlm.nih.gov/41890060). bioRxiv : the preprint server for biology, 2026. - [Chromosome-level haplotype-resolved assembly of highly heterozygous grass genomes with PhaseGrass.](https://doi.org/10.1038/s41467-025-66377-5). 2025. - [Accessing medically relevant complex regions with a pangenome graph of 20 near-complete Japanese haplotypes.](https://doi.org/10.1038/s41467-026-73461-x). 2026. - [Haplotype-resolved genome of Citrus × sinensis 'Pera IAC', the most widely cultivated sweet orange in Brazil.](https://doi.org/10.1038/s41597-026-07229-9). 2026. - [A chromosome-scale reference genome and integrative transcriptome provide insight into tissue- and stress-specific responses in tetraploid sainfoin (Onobrychis viciifolia).](https://doi.org/10.1007/s00425-026-05021-y). 2026. > This article is educational and does not replace validated analysis plans, institutional policy, clinical interpretation, or specialist review.