This is a repository describing quality control methods and tools to assess genome assemblies.
Correctness is the base level accuracy of the assembly. The more correct the assembly, the fewer single nucleotide polymorphisms (SNPs) or insertions/deletions (INDELs) it will have.
Approaches to test correctness:
- K-mer fequency based approaches (e.g. yak and merqury) (this may need short reads - determine)
- Frameshift detection using transcript data
- N50/NG50
- QUAST Nx/NGx
BUSCO shows the presence of conserved single copy genes present in an evolutionary group. This is a standard and useful approach for assessing the quality of a genome assembly. However it has its limitations, we observed some of these in O. tsutsugamushi. Orientia spp. have lost 8% of the single copy genes present in the rest of rickettsiales and the database. Another limitation observed was that assemblies could score very highly in BUSCO, yet still be highly fragmented. This is because the repetitive sequences, which lack the conserved single copy genes, are challenging to assemble.
CheckM provides a similar function to BUSCO. However, it relies on marker genes present in the taxonomic group. Unlike with BUSCO this enables the detection of contamination.
In most cases of de novo genome assembly there is no reference availible for comparison. However, occasionally when developing new sequencing/assembly methods de novo assemblies can be compared to a reference genome. This makes quality assessments much simpler.
Alignment of the contigs/genome assembly to the reference is the simplest and best way to assess the success. For bacterial genomes, it is important to have genomes/contigs correctly rotated for alignment (e.g. using circlator. Alignment can then be performed by tools like minimap2 to output a sorted indexed bam file.
Visualisation of these bam files can be done using tools like mauve or IGV.
Ryan Wick developed a script, compare_assemblies.py, for comparing different assemblies (initially intended for assemblies pre-polishing and post-polishing). However, when assemblies are structurally very similar to the reference, it is possible to use this script to identify SNPs and INDELs found in the new assembly.
However, one challenge can be identifying which of these errors are true positives and which are false positives (i.e. wrong in the reference). One way to deal with this is to examine the evidence for these at a read level.
A fast but lower resolution method to compare these assemblies is using dotplots, performed in tools like gepard.