Lesson 8 of 9
Module 6 · Genome assembly
A genome becomes useful to others when it is deposited in a public archive with clear names, annotation and a description of how it was made.
Before submission
- Remove very short contigs (most archives require at least 200 bases) and any contamination.
- Rename contigs simply and consistently, for example contig_1, contig_2, rather than assembler-specific names.
- Circular chromosomes and plasmids are usually rotated to start at a standard gene, such as dnaA for bacterial chromosomes.
- Annotation goes in a GFF3 file; archives can also annotate bacterial genomes for you.
Why each step
- Short contigs rarely hold a whole gene and are more likely to be errors or fragments of contamination. Genes on your assembly: a first count, and is it complete? showed how GC content, coverage and duplicated marker genes reveal contamination; anything flagged there should be removed before submission, not after.
- Names such as NODE_1_length_7294_cov_30.961044 mix the name with statistics that mean nothing outside SPAdes. Simple names are easier to cite and are kept unchanged by other tools.
- Rotation. A circle has no natural first base, so an assembler may start a circular chromosome anywhere. The dnaA gene, which encodes the protein that starts DNA copying, sits next to the chromosome's origin of replication, so starting every assembly there gives genomes of the same species comparable coordinates.
- GFF3 (General Feature Format, version 3) is a tab-separated text file with one line per feature, such as a gene, giving the contig it is on, its start and end positions, its strand and its attributes, such as a name.
The first decision comes before any of these steps: which assembly to share. In this module you made two, the five-contig short-read assembly from Short-read assembly with SPAdes and the single long-read contig from Long reads and repeats. Once polished, the long-read assembly is the better record: it is nearly complete, holds every copy of the repeat in its place, and after short-read polishing its bases are accurate too. Archives record how finished an assembly is, using levels from contig and scaffold up to chromosome and complete genome, so state honestly what you have. A draft is a perfectly good submission; a draft labelled complete is not.
seqkit seq -m 200 spades_out/contigs.fasta \
| seqkit replace -p '.+' -r 'contig_{nr}' > practice_bacterium_assembly.fasta
grep ">" practice_bacterium_assembly.fastaWhat this does
seqkit seq -m 200reads the SPAdes contigs and passes on only sequences of at least 200 bases (-msets the minimum length).- The backslash continues the command on the next line, and the pipe
|sends the filtered contigs straight into the next command without writing an intermediate file. seqkit replacechanges each sequence's name.-p '.+'is the pattern to find: in this pattern language a dot means "any character" and a plus means "one or more", so it matches the whole name.-r 'contig_{nr}'is the replacement, where{nr}is the record's number in the file: 1, 2, 3 and so on.> practice_bacterium_assembly.fastasaves the result in a new file, leaving the original SPAdes output untouched.- The last line prints only the header lines of the new file, as in Short-read assembly with SPAdes.
Reading the output
There are still five sequences, so none was shorter than 200 bases; the shortest, the repeat contig, is 2,001 bases. They keep their order, longest first, so contig_1 is the old NODE_1 and contig_5 the old NODE_5. Note the coverage information is gone from the names. If you need it later, keep the original file, or record the old and new names side by side.
A quick check after any filtering step is to run seqkit stats on the new file, as in What genome assembly is, and compare the number of sequences and total length with the original. If a filter removed far more than you expected, for example because the minimum length was typed as 2000 instead of 200, the totals show it at once. The same habit applies to every file you change before submission: compare before and after, and keep the original.
Go deeper: the overlap at the ends of a circle
When an assembler closes a circular chromosome, it does not always cut the circle cleanly. Some leave the same short stretch, from tens to thousands of bases depending on the assembler, at both the start and the end of the contig, because the last reads overlap the first. Submitted as it is, the genome is too long by that amount, and any gene in the overlap appears twice.
So the order of steps matters. First confirm that the contig really is circular and trim the duplicated end; then rotate it to start at dnaA, on the strand where dnaA reads forwards. Rotating first would move the overlap into the middle of the chromosome, where it is much harder to spot. Tools such as Circlator were written to automate these checks, but it is worth confirming the result by eye.
Where genomes go
The International Nucleotide Sequence Database Collaboration links three archives: NCBI GenBank (USA), ENA (Europe) and DDBJ (Japan). Submit to one and it appears in all three. A submission links a BioProject (the study), a BioSample (the organism and where it came from), the raw reads, and the assembly.
The collaboration is usually shortened to INSDC. Its three members exchange their records, so a scientist in any country can find your genome in whichever archive they use. Each record receives an accession, a permanent identifier you quote in your paper. BioProject accessions begin with PRJ and BioSample accessions with SAM, so you can tell at a glance what kind of record an identifier points to.
The raw reads are deposited in the archives' read databases, NCBI's Sequence Read Archive or ENA's equivalent, and get accessions of their own, linked to the same BioSample. Archives also check what you send. Submitted assemblies are screened for adapter and for sequence that looks as if it comes from another organism, and flagged contigs are returned for you to trim or remove. Running the checks of Genes on your assembly before submitting saves a round of corrections, and it is better to find a contaminant yourself than to have it found for you.
The BioSample is where most of the useful description lives: the species and strain, and details such as where and when the sample was collected and from what source, for example a patient, soil or a river. Archives ask for a minimum set of these fields for each kind of sample. Filling them in carefully is what lets someone years later find every isolate of a species from one region, or link your genome to an outbreak.
Worked example · a methods sentence
Step by step. Each part of the sentence comes from a lesson in this module. "Paired 125-base reads" and "59× coverage" come from the seqkit output and the calculation in What genome assembly is (1,875,000 ÷ 31,950 ≈ 59). "--isolate" is the SPAdes setting from Short-read assembly with SPAdes. The long-read coverage comes from Long reads and repeats: 959,152 bases ÷ 31,950 ≈ 30. "--nano-hq" is the Flye setting used there. The blanks after "v" are version numbers, which you fill in from the tools you actually ran; the blank after "polished with" is the polishing tool, or tools, you used.
What the answer means. A good methods sentence lets another person repeat your work exactly: which data, which tool, which version, which settings. The blanks are deliberate. A version number written from memory, or copied from someone else's paper, is a common and avoidable error; ask each tool for its version in the environment you used and copy that.
The other half: data availability. The methods say how the genome was made; a data availability statement says where it is. Once the archive has issued accessions, quote the BioProject, the BioSample, the read accessions and the assembly accession in the paper. Readers can then go straight from your methods sentence to the exact reads and contigs behind it, and repeat or extend the work.
Beyond bacteria: chromosomes, complete genomes and pangenomes
Scaffolding to chromosomes. A large plant or animal genome assembled from long reads usually still comes out in hundreds or thousands of contigs. Joining them into whole chromosomes is scaffolding, the same idea as the read-pair scaffolds of Short-read assembly with SPAdes at a far larger scale. The most common evidence today is Hi-C. In this laboratory method the DNA is fixed inside the nucleus, so that stretches lying close together in three dimensions are joined to each other, and the joined pairs are then read with short reads. Two stretches on the same chromosome touch far more often than stretches on different chromosomes, and near neighbours touch most of all. The contact counts therefore show which contigs belong to one chromosome, and in what order and orientation. Tools such as YaHS or 3D-DNA build the scaffolds, and the result is usually checked by eye on a contact map and corrected by hand. Optical maps, images of very long single DNA molecules labelled at a short sequence motif, are another way to order contigs and to check for wrong joins.
Telomere-to-telomere genomes. A telomere is the protective repeated sequence at each end of a chromosome. A telomere-to-telomere (T2T) assembly has every chromosome in one gap-free piece, from one telomere to the other. The first complete human genome, known as T2T-CHM13, was published in 2022. It added nearly 200 million bases, about 8% of the genome, that the previous reference had lacked, most of it in centromeres and other long repeats, and it relied on HiFi and ultra-long nanopore reads together (Long reads and repeats). It came from a cell line that carries essentially one version of each chromosome, which removed the problem of telling the two parental copies apart. Its Y chromosome was added in 2023, from a different person. Complete genomes have since been reported for other species, including crops such as rice and maize.
The one-version cell line mattered because of phasing. Each of us carries two versions of every chromosome, one from each parent, and each version is called a haplotype. Where the two differ, an assembler sees a bubble in the graph, as in Assembly graphs, and must either pick one route, mixing the parents' sequence, or keep both apart. Keeping them apart is phasing. Assemblers such as hifiasm, from Long reads and repeats, can now do this for ordinary diploid samples, especially with help from Hi-C or from the parents' reads, which is what made the many two-haplotype assemblies of the human pangenome possible.
Pangenomes and graph references. One genome, even a complete one, is still one individual. A pangenome brings together genomes from many individuals of a species, so that the variation between them becomes part of the reference. In 2023 the first draft of a human pangenome reference was published: assemblies of both chromosome copies from 47 people of diverse ancestries. Such a reference is stored as a graph, shown in the figure below: shared sequence is a single path, and where people differ the graph splits into alternatives. Lining reads up against a graph lets each person's DNA be compared with the version closest to their own, which reduces the bias of a single reference. Minigraph-Cactus and PGGB are examples of tools that build such graphs. In bacteria the word pangenome is also used for the full set of genes found across all strains of a species, of which only a part, the core genome, is shared by every strain.
How to read the figure below
The top panel is a linear reference, one line of sequence. The bottom panel is a small graph built from several genomes. Dark boxes are sequence that all the genomes share; the orange pair is a place where some genomes have G and others T; the pink box is extra sequence that some genomes carry and the lower arrow lets the others skip. Follow any route from left to right and you spell one genome.
The complete human genome is Nurk et al. (2022, Science); the complete Y chromosome is Rhie et al. (2023, Nature); and the draft human pangenome reference is Liao et al. (2023, Nature). Hi-C was introduced by Lieberman-Aiden et al. (2009, Science). The three archives are described in Nucleic Acids Research, for example Arita, Karsch-Mizrachi and Cochrane (2021) for the INSDC. When you use a reference, cite its exact version: T2T-CHM13 v2.0 adds the Y chromosome from a different person, HG002, which v1.1 lacks, so results on the two are not directly comparable.
Key point
Deposit the raw reads as well as the assembly. Assemblers improve, and others may want to reassemble your data.
Reads are also useful for questions you did not ask. A resistance gene described for the first time years after your study can be searched for directly in the deposited reads of thousands of old isolates, including in any part of a genome that an assembly left out or collapsed, such as the repeat in this module.
Further reading · Module 6ENA: submitting a genome assembly. · NCBI: submitting a genome to GenBank. · The T2T-CHM13 project page: the complete human genome and its downloads. · The Human Pangenome Reference Consortium: the pangenome reference and its data. · All reading for this course
Check your understanding
1. You submit to ENA. Where else will the genome appear?
2. What does seqkit seq -m 200 do?
3. Why deposit raw reads as well as the assembly?
Up next
Project
Coverage and contiguity