Genome Polishing and Consensus Correction

  • Genome polishing and consensus correction are important stages of genome assembly that improve the accuracy of assembled DNA sequences by identifying and correcting errors remaining after the initial assembly process. Genome assembly reconstructs genomic sequences from sequencing reads, but the resulting contigs and scaffolds may contain incorrect bases, small insertions, deletions, substitutions, and local alignment errors. These errors can arise from sequencing technology limitations, insufficient coverage, repetitive regions, incorrect read overlaps, and limitations of assembly algorithms. Polishing uses the original sequencing data and computational methods to identify likely errors and produce a more accurate consensus sequence.
  • A genome assembly is not necessarily error-free simply because it contains long contigs or chromosome-scale scaffolds. Assembly continuity and nucleotide accuracy are different measures of quality. A highly continuous assembly may still contain numerous incorrect bases, while a fragmented assembly may have relatively accurate sequences within its contigs. Genome polishing focuses primarily on improving sequence-level accuracy, whereas scaffolding and gap filling focus on connecting contigs and reconstructing missing regions. These processes complement one another and are often performed at different stages of a genome assembly workflow.
  • Consensus correction is based on the principle that multiple sequencing observations can provide evidence for the most likely nucleotide at a particular genomic position. When reads from the original sequencing experiment are aligned to an assembled sequence, the nucleotide observations can be compared with the assembly. If several high-quality reads consistently support a different nucleotide from the assembled base, the position may contain an assembly error. A consensus algorithm evaluates the evidence and determines whether the existing base should be retained, replaced, inserted, or deleted. The resulting sequence is a consensus representation supported by the available reads and the assumptions of the correction method.
  • Sequencing errors differ among technologies and platforms. Illumina short-read sequencing generally produces highly accurate reads, but errors can still occur because of base-calling limitations, low-quality cycles, amplification artifacts, and systematic sequence-context effects. PacBio continuous long reads historically had relatively high raw error rates, although modern circular consensus sequencing produces highly accurate HiFi reads. Oxford Nanopore sequencing can generate very long reads, but its error profile depends on the sequencing chemistry, pore technology, basecaller, and software version. Errors may include substitutions, insertions, and deletions, and some sequence contexts can be more difficult to resolve than others.
  • Genome polishing methods are therefore selected according to the sequencing technology used for assembly and the available supporting reads. Short-read polishing uses accurate short reads to correct errors in an assembled sequence. Long-read polishing uses long reads to improve continuity and correct errors in long-read assemblies. Hybrid polishing combines evidence from different sequencing technologies. Multiple rounds of polishing may be performed, but repeated correction is not automatically beneficial. Excessive polishing, unsuitable parameters, or incorrect read mapping can introduce new errors or remove genuine biological variation.
  • Short-read polishing commonly uses Illumina reads aligned to the assembled genome. Because short reads generally have high per-base accuracy, they can provide strong evidence for correcting substitutions and small insertions or deletions. However, short reads may map ambiguously in repetitive regions, duplicated genes, and other complex genomic areas. If reads from multiple repeat copies are incorrectly mapped to a single assembly location, the polishing process may introduce incorrect bases. Short-read polishing should therefore use appropriate mapping parameters, read-quality filtering, and careful interpretation of ambiguous alignments.
  • Pilon is a widely used tool for improving genome assemblies using aligned short-read data. It can identify and correct several types of sequence errors, including base substitutions, small insertions and deletions, and some local assembly problems. It can also provide information about gaps and possible local structural differences, depending on the input evidence and software version. Pilon is commonly used with bacterial and other relatively compact genome assemblies, although its practical suitability depends on the dataset and computational requirements. It requires an assembly and read alignments in an appropriate format.
  • Polypolish is another short-read polishing tool designed to improve assemblies while addressing some limitations associated with mapping ambiguity, particularly in repetitive regions. It uses alignments from short reads to identify likely corrections and can be useful for bacterial genome assemblies. Its approach is intended to reduce errors caused by reads being incorrectly assigned to repeated genomic regions. The effectiveness of Polypolish depends on the quality of the assembly, the read data, and the alignment evidence.
  • For long-read assemblies, polishing methods are designed to address the error characteristics of the sequencing platform. PacBio HiFi reads already have high consensus accuracy, so extensive additional polishing may not always be necessary. However, assembly-specific errors can remain, particularly in difficult repetitive or structurally complex regions. Oxford Nanopore assemblies may benefit from consensus correction using reads generated with the same or a compatible sequencing chemistry and basecalling workflow. The correction model must match the characteristics of the input data because different chemistries and basecalling models can produce different error profiles.
  • Racon is a consensus-correction tool commonly used with long-read sequencing data. It improves an assembly by using alignments between reads and the draft sequence to generate a corrected consensus. It is often used after a long-read assembler has produced an initial assembly. Racon can be applied iteratively, with each round using updated alignments against the latest consensus. The number of rounds should be determined by evaluating the improvement in sequence accuracy rather than applying an arbitrary number of corrections.
  • Medaka is a consensus-polishing tool developed for Oxford Nanopore sequencing data. It uses neural-network models to estimate an improved consensus from read alignments and the draft assembly. The appropriate Medaka model depends on the sequencing chemistry and basecalling configuration. Using an incompatible model can reduce accuracy rather than improve it. Medaka is generally used after reads have been aligned to the draft assembly, and its results should be evaluated using independent quality metrics and, where possible, additional sequencing evidence.
  • Arrow is a consensus-polishing approach associated with PacBio sequencing workflows. It uses PacBio read evidence to improve the consensus sequence and has been used in workflows involving PacBio continuous long reads and HiFi-related data. Current PacBio software and recommended polishing procedures depend on the data type and platform version. Researchers should use the appropriate current PacBio tools and models for their particular sequencing chemistry rather than assuming that one polishing program is suitable for all PacBio datasets.
  • Hybrid polishing combines short-read and long-read evidence to improve different aspects of sequence accuracy. Long reads provide information across large genomic regions and can help resolve structural ambiguities, while accurate short reads can provide strong evidence for local base correction. A common strategy is to assemble a genome using long reads, correct the assembly using long-read consensus methods, and then apply short-read polishing when suitable Illumina data are available. The order and choice of tools depend on the assembly method, read quality, genome complexity, and whether the assembly represents a haploid sequence, a diploid genome, or multiple haplotypes.
  • Before polishing, sequencing reads should undergo appropriate quality control and preprocessing. Low-quality reads, adapter contamination, sequencing artifacts, and contaminating sequences can affect the correction process. Read quality can be assessed using FastQC and summarized with MultiQC. Tools such as fastp and Cutadapt can be used for short-read preprocessing, while long-read data may require platform-specific filtering and quality assessment. Researchers should preserve the original sequencing data and document all preprocessing decisions so that the polishing process can be reproduced.
  • Read alignment is a central step in most polishing workflows. Short reads are commonly aligned using BWA-MEM or Bowtie2, while long reads are often aligned using minimap2. The aligner and its parameters should be appropriate for the sequencing technology and the intended analysis. For example, minimap2 provides different presets for Oxford Nanopore reads, PacBio HiFi reads, and other long-read data types. Incorrect alignment settings can lead to misplaced reads, particularly in repetitive regions, and may cause the polishing tool to make inappropriate corrections.
  • The following example demonstrates how Illumina paired-end reads can be aligned to a draft genome using BWA-MEM and prepared for short-read polishing:

bwa index draft_assembly.fasta

bwa mem -t 16 draft_assembly.fasta \

    reads_R1.fastq.gz reads_R2.fastq.gz \

    > reads.sam

samtools view -@ 8 -bS reads.sam \

    | samtools sort -@ 8 -o reads.sorted.bam

samtools index reads.sorted.bam

  • The resulting sorted and indexed BAM file contains the short-read alignments required by compatible polishing tools. The alignment should be examined for mapping quality, coverage, and unexpected patterns before correction. Reads that map to multiple locations may provide ambiguous evidence, particularly in repetitive or duplicated regions.
  • Pilon can be used to polish an assembly using the aligned short reads. A simplified example is:

java -Xmx16G -jar pilon.jar \

    –genome draft_assembly.fasta \

    –frags reads.sorted.bam \

    –output polished_assembly \

    –outdir pilon_output \

    –threads 16

  • The memory and thread settings should be adjusted according to the available computational resources. Pilon may produce corrected assembly sequences and additional output files describing changes and potential assembly problems. Its documentation should be consulted for supported options and input requirements. In some workflows, the corrected assembly is used for another alignment and polishing round, but additional rounds should only be performed when they provide measurable improvement.
  • Long-read polishing can be performed using minimap2 to align reads to the draft assembly, followed by Racon to generate a corrected consensus. For Oxford Nanopore reads, an illustrative workflow is:

minimap2 -t 16 -ax map-ont \

    draft_assembly.fasta long_reads.fastq.gz \

    | samtools sort -@ 8 -o long_reads.sorted.bam

samtools index long_reads.sorted.bam

racon -t 16 \

    long_reads.fastq.gz \

    long_reads.sorted.bam \

    draft_assembly.fasta \

    > racon_polished.fasta

  • The exact input and alignment requirements should be checked against the installed Racon version. For successive polishing rounds, reads should be realigned to the latest corrected assembly rather than reusing alignments against an outdated sequence. The same principle applies to other consensus-correction workflows.
  • For Oxford Nanopore assemblies, Medaka can be used after generating an appropriate read alignment or supplying the required reads and draft sequence through its supported workflow. A simplified example using a BAM file is:

medaka_consensus \

    -i long_reads.fastq.gz \

    -d racon_polished.fasta \

    -o medaka_output \

    -t 16 \

    -m appropriate_model

  • The model name shown here is a placeholder and must be replaced with a model compatible with the actual sequencing chemistry and basecalling configuration. The available model names and recommended commands depend on the installed Medaka release. Users should consult the tool documentation before running the command.
  • Polishing can correct several types of sequence errors. Substitution errors occur when an incorrect nucleotide is present in the assembly. Insertion errors occur when one or more extra bases have been included, while deletion errors occur when bases that should be present are missing. Small insertion and deletion errors are particularly important in protein-coding regions because they can cause frameshifts, altering the reading frame and potentially changing the predicted protein sequence. Polishing can therefore improve gene prediction and functional annotation by restoring more accurate coding sequences.
  • Consensus correction can also improve the identification of single-nucleotide variants, small insertions and deletions, regulatory regions, and other genomic features. However, polishing is not a substitute for variant calling or biological validation. The goal of polishing is generally to produce a representative consensus sequence for the intended assembly, not to identify every difference among individuals or haplotypes. In diploid and polyploid genomes, genuine differences between homologous chromosomes must be preserved when the goal is to produce haplotype-resolved assemblies. A polishing workflow that treats real heterozygous variation as an error may incorrectly remove biological information.
  • A major challenge in polishing is distinguishing technical errors from genuine biological variation. If an assembly contains a heterozygous position, reads from different haplotypes may support different nucleotides. A consensus algorithm may select the most frequent base and remove the alternative allele, even when both alleles are biologically real. Similar problems can occur in polyploid organisms, mixed microbial populations, and metagenomic datasets. Haplotype-aware polishing and careful read assignment are important when preserving biological variation is necessary.
  • Repetitive sequences also complicate consensus correction. Reads from multiple repeat copies may align to the same assembly region, producing misleading support for a particular nucleotide or indel. A collapsed repeat may contain a mixture of reads from several genomic copies, and polishing may generate a consensus that does not accurately represent any individual copy. In these regions, long reads spanning the repeat and its unique flanking sequences, together with assembly graph information, can provide more reliable evidence. Polishing should not be used to conceal unresolved structural ambiguity.
  • Coverage is another important consideration. Low coverage provides limited evidence for determining the correct base, while extremely high or uneven coverage may indicate repetitive regions, copy-number variation, contamination, or amplification bias. Consensus tools use different statistical or machine-learning models to interpret read support, and their assumptions may not be appropriate for every dataset. Researchers should examine coverage distributions and alignment quality rather than assuming that the most frequently observed base is always correct.
  • Genome polishing should be evaluated using independent and complementary methods. One approach is to map the original sequencing reads back to the polished assembly and examine mapping rates, coverage, mismatch rates, and insertion or deletion patterns. Improved alignment consistency can indicate that the consensus has become more compatible with the read data. However, read mapping alone does not prove correctness because the same reads were used to perform the polishing. Independent validation data, when available, provide stronger evidence.
  • K-mer-based evaluation is another useful method. A k-mer is a short sequence of length k extracted from a genome or sequencing reads. Tools such as Merqury compare k-mers in the sequencing data with those in the assembly to estimate consensus quality and completeness. These analyses can identify missing or unsupported sequence content and provide a complementary assessment of nucleotide accuracy. The interpretation depends on sequencing coverage, read quality, genome heterozygosity, and whether the k-mer set represents the intended haplotype or genome.
  • QUAST can provide assembly statistics and support comparisons between draft and polished assemblies. BUSCO can assess the completeness of conserved genes, although it is not a direct measure of nucleotide-level consensus accuracy. For microbial genomes, additional validation may include comparison with closely related reference genomes, examination of conserved genes, and assessment of expected genome size and gene content. For large eukaryotic genomes, long-range evidence such as Hi-C, optical mapping, and genetic maps may be needed to evaluate structural correctness independently of base-level polishing.
  • A simplified QUAST command for evaluating a polished assembly is:

quast.py polished_assembly.fasta \

    -o quast_results \

    -t 16

  • The resulting report can be compared with the report from the original assembly. Researchers should examine total assembly length, number of contigs and scaffolds, N50, L50, and other relevant statistics. Polishing may not substantially change N50 because its main purpose is to correct sequence-level errors rather than increase contiguity. A decrease in assembly length or a change in contig structure should be investigated to determine whether it reflects valid correction or an unintended loss of sequence.
  • Genome polishing has important applications in microbial genomics, human genomics, plant and animal genome projects, comparative genomics, metagenomics, and evolutionary research. In microbial genomics, accurate polished assemblies can improve the identification of genes, antimicrobial resistance determinants, plasmids, and genomic islands. In human and animal genomics, polishing can improve gene annotation, variant interpretation, and the analysis of structural and functional genomic regions. In plant genomics, it can improve the reconstruction of large, repetitive, and often polyploid genomes. In metagenomics, polishing can improve metagenome-assembled genomes, although strain mixtures and closely related organisms make consensus correction particularly challenging.
  • Genome polishing also has limitations. It cannot reliably reconstruct large missing genomic regions without suitable supporting evidence, and it does not automatically resolve incorrect scaffold joins or complex structural misassemblies. It may introduce errors when reads are mapped incorrectly, when the wrong model is used, or when biological variation is mistaken for technical noise. Repeated polishing rounds may produce diminishing returns and can sometimes reduce accuracy. The best strategy is to select polishing methods that match the sequencing technology, assess the quality of the input data, and evaluate the results after each major correction stage.
  • For reproducible genome assembly, researchers should preserve the original draft assembly, raw sequencing reads, quality-control reports, read alignments, polished assemblies, and all relevant software versions and parameters. They should document the sequencing platforms, library preparation methods, basecalling software, reference genomes, and polishing tools used. The number and order of polishing rounds should be reported, along with the metrics used to assess improvement. When possible, independent sequencing data or orthogonal technologies should be used to validate important corrections.
  • Genome polishing and consensus correction are essential for producing reliable genome assemblies. They improve nucleotide-level accuracy by using sequencing evidence to correct substitutions, insertions, deletions, and other local errors. Short-read, long-read, and hybrid polishing approaches provide different advantages, and the appropriate strategy depends on the sequencing technology, genome complexity, and biological objectives. When combined with careful read alignment, haplotype-aware analysis, independent validation, and reproducible documentation, genome polishing helps transform draft assemblies into more accurate genomic resources for research and practical applications.
Author:

Leave a Reply

Your email address will not be published. Required fields are marked *