A field guide to genome curation
Hi-C contact maps reveal the ordering and orientation errors in a draft genome assembly. This guide explains how to read the map, recognize the common misassembly signatures, and correct them in OpenPretext, a contact map viewer that runs in the browser.
Reference genomes are curated by hand.
Assemblers join sequencing reads into longer pieces, but they make errors: they join sequences that do not belong together, place fragments in the wrong orientation, and leave chromosomes in scattered pieces. A Hi-C contact map makes these errors visible, because DNA that sits close together in the nucleus produces a stronger signal and a correctly assembled chromosome produces a distinctive pattern.
Correcting these patterns by hand is called curation. The expertise has been concentrated in a small number of large sequencing centers. OpenPretext runs the same workflow in a web browser at no cost, and this guide and its tutorials are intended to teach the process, so that a lab with a new genome can curate it without specialist support.
How to read a contact map
Curation rests on one property of Hi-C data: sequences that are physically close in the nucleus produce more contacts and appear brighter on the map. Four features carry most of what you need to read.
The bright line down the middle
Every position contacts itself most of all, so the main diagonal is always the strongest signal. It is the baseline for reading everything else.
Squares on the diagonal are chromosomes
A chromosome contacts itself far more than anything else, forming a bright square. Clean squares with sharp edges mean a well assembled chromosome.
Signal fades away from the diagonal
The farther apart two positions are, the fewer contacts they share. Bright signal far from the diagonal usually means a sequence is in the wrong place.
Tracks along the edges are your aids
Coverage, gaps, telomere density, and repeat density run as tracks down the sides of the map, read from the file or added from a reference or coverage data you load. A coverage dip or a gap marks where to look; telomere and repeat peaks mark real biology, not errors. Press X to toggle them.
Four errors to fix, and two things to leave alone
Assembly errors produce the same recognizable patterns across genomes, so each fix is a matter of identifying the pattern and applying the right operation. Each is shown here next to a correctly assembled chromosome for comparison. Misordered fragments and inversions are the most common; translocations are rarer but the most disruptive to a final assembly.
A finished chromosome
One clean, bright square. A strong continuous diagonal, sharp edges, and little signal outside the block.
A piece assembled backward
The diagonal breaks and crosses itself into an X or bowtie. A stretch of sequence is present but pointing the wrong way.
Two chromosomes wrongly joined
A bright rectangle sits far from the diagonal, linking regions that should not touch. The order of pieces is wrong.
One contig, two chromosomes
Inside a single piece the signal drops sharply and restarts. The assembler fused two unrelated regions into one contig.
A chromosome in many parts
Lots of small blocks with faint links between them. The pieces belong together but sit out of order, common for small chromosomes.
A centromere
A faint band across an otherwise clean chromosome. Repetitive regions like centromeres look unusual but are real biology. Do not cut them.
Real maps are noisier than the diagrams
The diagrams above are idealized so that each pattern is clear. Below is a real bird genome as it comes from the assembler, before any curation.
King quail, 645 contigs, before curation. OpenPretext overview.
The strong line down the middle is the same diagonal described above. The red marks off that diagonal are misassemblies: the short crossing streaks are inversions, the blocks sitting away from the center are pieces in the wrong place, and the dense band is a cluster of small chromosomes that still need ordering.
Each mark corresponds to one of the four signatures, and each has a single-key operation to correct it. Curation reduces this map to the clean set of squares shown at the start.
Getting your data into OpenPretext
To follow the tutorials or explore the tool you need nothing of your own: Load Example downloads a real assembly, and the tutorials run on real data. To curate your own genome, OpenPretext opens a .pretext file, the same contact-map format PretextView reads. If you already have one, open it and start. If not, you build one from your draft assembly and its Hi-C reads in three steps.
Align the reads
Map the paired Hi-C reads against your draft assembly with a short-read aligner such as bwa-mem2, bwa mem, or a dedicated Hi-C mapping pipeline. This produces a BAM of alignments.
In: assembly FASTA and Hi-C read pairs.Build the contact map
Pipe the alignments through PretextMap, which bins the contacts and writes a compact .pretext file. Coverage, gap, and telomere tracks can be embedded afterward with PretextGraph.
Out: one .pretext file.Open it
Drag the .pretext onto OpenPretext, or use Open. The loader reads the coarse overview first, so even a large map opens quickly in the browser.
Ready to curate.samtools view -h aligned.bam | PretextMap -o assembly.pretext
The Pretext tools come from the Wellcome Sanger Institute: PretextMap builds the map and PretextGraph embeds the annotation tracks. For ready-made assemblies to practice on, GenomeArk hosts public genome data, including the Hi-C maps behind Load Example.
From draft to finished assembly
Curation is a loop repeated until the map is clean. OpenPretext provides a tool for each step, along with automated help and a score that indicates when the assembly is finished.
Inspect
Scan the whole genome and mark the signatures: inversions, wrong joins, chimeras, scattered pieces. Turn on the overlay tracks (coverage, gaps, telomeres) as you scan; the app also flags contig junctions the Hi-C signal does not support and contigs that look like retained haplotigs.
Locate errors before correcting them.Cut and join
Split chimeric contigs at the break; merge pieces that belong together.
Press E for edit mode, then C to cut and J to join.Flip and order
Reverse backward pieces and drag contigs into the right sequence so the diagonal strengthens.
F to flip.Assign scaffolds
Group the ordered pieces into chromosomes. Let auto-sort and auto-assign scaffolds do the first pass.
Review the automated grouping before accepting it.Check the health score
A single number rates contiguity, decay, and structure, and it rises as the assembly improves. The before/after map puts the assembly as loaded beside the one your curation produced, so you can compare the assembly you started from with the one you produced.
The score rises as the assembly improves.Export
Save the curated assembly as AGP or FASTA, with every reorder and flip applied. The full record of what you changed is kept in the saved session, which can also be exported as a replayable script. An AGP can be imported again later to resume a curation in progress.
Output for downstream analysis.Telling biology from error
The harder half of curation is not finding the errors but knowing what to leave alone. Several features of a real genome make the same shapes on the contact map as a misassembly, and cutting them breaks a correct assembly. You have already met the clearest case: a centromere is a faint band across a chromosome, not a break, and it stays. Three harder calls follow.
Satellite and rDNA arrays
Highly repetitive regions such as satellite DNA and ribosomal RNA arrays make a dense, speckled band with faint off-diagonal specks, because reads from the repeats map to more than one place. The pattern can resemble a chimeric break or a scatter of small errors.
A retained alternate haplotype
A retained haplotig and a real segmental duplication both appear as a bright block off the diagonal, where one region contacts a homologous region elsewhere. On the map alone the two are indistinguishable.
A degenerate sex chromosome
A degenerate sex chromosome, the W in birds or the Y in mammals, is gene-poor and repeat-rich, so its block is faint and sparse next to the autosomes. It can read as a low-quality or half-assembled region.
Three questions settle most of these calls. Does the feature line up with a track, coverage, repeats, or telomeres, that marks it as real? Does the automated detection flag it, or leave it alone? And does the edit you are considering actually help: a correct cut makes the diagonal cleaner and moves the metrics up, while cutting real biology does neither. When you are unsure, zoom in and let the signal decide before you commit.
The same map supports further analysis
Once an assembly is clean, the same contact map can be used to measure genome structure. These analyses are optional and sit outside the core curation workflow.
Structure readouts from the map
Three measurements taken directly from the map, with no additional experiments: insulation scores (where chromosome neighborhoods begin and end), P(s) decay (how tightly the genome is packed), and A and B compartments (active versus quiet regions). This is an emerging area, and the concepts recur in the analysis page.
Shared curation strategies
OpenPretext's AI assistant runs on prompt strategies that curators write and share, tuned for inversions, fragmented assemblies, micro-chromosomes, and other cases. The collection is open for contributions.
Start with the interactive tutorials
The interactive tutorials load a real genome and work through each step, from reading the map to exporting a finished assembly. OpenPretext is free, open source, and runs without installation.