Repository navigation
Example
A small simulated dataset, bundled with BACoN, to check an installation and to see every output in a few
seconds. example/make_example.py
generates it (standard library only; the same files every time):
- a 30 kb circular reference ("organelle");
- four samples of about 50x of simulated Nanopore reads (1–20 kb, 1.5% errors), each mixed with 20% of reads from an unrelated sequence that baiting must remove;
- known SNPs: alpha is identical to the reference, beta has 6 SNPs, gamma has beta's 6 and 4 more, delta has 10 of its own.
| Reference | alpha | beta | gamma | delta | |
|---|---|---|---|---|---|
| Reference | 0 | 0 | 6 | 10 | 10 |
| alpha | 0 | 0 | 6 | 10 | 10 |
| beta | 6 | 6 | 0 | 4 | 16 |
| gamma | 10 | 10 | 4 | 0 | 20 |
| delta | 10 | 10 | 16 | 20 | 0 |
conda activate BACoN
bash example/run_example.sh example_output # optional: number of threads as a second argumentThe script generates the data in example_output/data/, runs BACoN with the default settings into
example_output/bacon/, and compares the SNP distances with the table above:
OK: all 6 pairwise SNP distances match the truth (example_output/bacon/4_compared/ska/snp_distances.tsv)
Open example_output/bacon/report.html in a browser. The report of this example is also online:
view it
or download it.
(Local paths were removed from it before publication: the program paths start with $CONDA_PREFIX.)
Overview and samples. All four samples assembled: about 250 of the 300 reads of each sample matched the
reference (the others are the off-target reads), and after filtering the depth is about 48x. The templated
assembly is 30,000 bp for each sample, as long as the reference, with no N base.

Tree. alpha sits with the reference; beta and gamma form a clade (their 6 shared SNPs); delta is on its own branch.

SNP distances. The heatmap, in tree order, shows the distances of the table above; the reference and alpha are listed as identical genomes.

Methods and run. The methods paragraph describes what was run, with the program versions; the run section records the command, the reference's MD5 and every program used.

For real data, see the Tutorial (28 potato cultivars) and its report.