Coalescent simulations of k-mer spectra to test Genome Profiling tools.
- Set up conda environment:
bash 00setUpEnv.sh - Run pipeline:
bash -i 01pipeline.sh - Adjust paprameters in
01pipeline.shand re-run if desired.
Simulating genetic data is not trivial. In the case of one diploid (i.e. two genomes), one can just throw in heterozygous sites according to the desired heterozygosity level (θ). If more than two genomes are simulated, the task becomes more complicated. This is because related genomes have a common ancestor and thus share mutations. A summary of the number of times different mutations are shared between genomes in a sample is called the site-frequency spectrum.
There are at the moment four steps that take place in a temporary directory. Only the final result, a k-mer spectrum, is stored in the working directory. The steps are:
- The simulation of genetic variants and storage of simulated genomes
- Generation of sequencing reads from the simulated genomes
- Generation of a KMC3 database
- Generation of a k-mer spectrum
Below, these steps are descibed in more detail.
This is done by the script makeGenomes.py, which runs the coalescent simulator msprime to generate genetic variant data in a sample of n genomes (where n is the ploidy level specified). msprime generates information only for variant sites and so the invariant rest of the genome is made up separately.
The genomes are then written (concatenated) to a single-record FASTA file, genomes.fa, with a specific header format containing the genome size information: >genomes_p[ploidy]_s[haploid GS]_t[total GS].
This script can be modified to allow for allopolyploidy, genomic repeats, unequal sub-genomes, etc.
This is done by makeReads.py, a very simple script that may well be replaced in the future by a more sophisticated sequnecing read simulator (there are several available for paired reads with error models).
For now, this generates error-free single-end reads in FASTA format which are written to reads.fa. To retrieve the genome size, the script looks at the header line of genomes.fa. It then generates reads of 150nt at 30x sequencing coverage. The read positions are sampled at random from a uniform distribution on the interval [1, genome size-150]. This should lead to narrow 'Poisson' peaks (bias parameter 0).
KMC3 is run on reads.fa with the k-mer length set to 21.
A k-mer spectrum is generated from the KMC DB and the temporary files are deleted.
Generating fabricated reference genomes from a template genome. If scaffolds are assigned to chromosomes, a tsv file can be provided together with a chromosome that should be samples (can be also A/X if just autosome vs sex chromosome needs to be distinguished). The newly sinthetised genome will either be out of windows stiched together (if -w is >0) or our of whole scaffolds (up to the size of desired generated genome). This is not a very efficient implementation, I tried to put something quick together. I recommend high -w for high quality reference genome templates.
For all options see
code/generate_reference_subset.py -h
The idea is that this way-out known genome can mimic some real-genome composition and other properties that would be hard to reproduce with any type of direct sequence simulation.