Review logic for drawing their locations as they appear to be pixel-imperfect and a little offset even at high DPI rendering.

It would be useful to expose certain Canvas methods or more easily pass in cartesian parameters via plot_stratton_spectrum. For example, passing in a predefined ymin and ymax so that multiple stratton spectrums will be on the same scale.

`canvas.cartesian(ymin=0, ymax=.12)' results in the new scale being layered on top of the old scale. By example:
test.pdf

sandialabs/toyplot#189

Install fails due to this.

Calling .density when there are no counts or probabilities means a divide by zero. Alert user with a custom ValueError exception.

Not sure if this is a problem with pip or an issue with snv-spectrum.

Using pip install snv-spectrum gives the following error: "Command "python setup.py egg_info" failed with error code 1 in /private/var/folders/mf/28775vz108sbnym4j9yd83w00000gn/T/pip-build-gypnrbfk/snv-spectrum/"

Looking at similar error codes suggests that it's due to an old version of setuptools, but I have 38.6.0 installed, so not likely the issue. Pip version 9.0.1.

https://github.com/astropy/pytest-remotedata

I'm running into situations in my work where I have two samples (say treatment and control) that seem like they have different trinucleotide signatures, but I want to rigorously show that to be the case. Some old literature suggests that a generalized form of Fischer's exact test may be the best way to proceed. I've attached the relevant literature to see if this is something you'd be interested in potentially implementing. I should point out that these papers, written in the 80's, make use of Monte Carlo approaches to approximating p-values. Given the large number of calculations needed to compute exact p-values, this approach is likely still valid. I also have some functional code, but it will need to be significantly cleaned up and better integrated into the Nucleic API.
agresti_wackerly_boyett.pdf
Adams&Skopek.pdf

Plot count or density heatmaps for a list of spectra with k=3 and identical reference_notation.

Figure 1A: Differences in mutation burden and spectra between carcinogen and genetic models¹

1 Westcott et al. (2014) The mutational landscapes of genetic and chemical models of Kras-driven lung cancer

>>> spectrum = Spectrum(3, 'purine')
>>> spectrum[Snv(reference='T', alternate='C', context='TTA')]
"""
~/miniconda3/lib/python3.6/site-packages/snv_spectrum-0.3.0-py3.6.egg/snv_spectrum/_snv_spectrum.py in __getitem__(self, key)
    209 
    210     def __getitem__(self, key):
--> 211         return self._substitutions[key]
    212 
    213     def __setitem__(self, key, item):

KeyError: Snv(reference="T", alternate="C", context="TTA")
"""

The solution is:

# Thymine is not a purine, it is a pyrimidine
>>> spectrum = Spectrum(3, 'purine')
>>> spectrum[Snv(reference='T', alternate='C', context='TTA').with_purine_reference]

Issue found by @scottrk2.

Python version: 3.6

Mes-MacBook-Air:snv-spectrum-master SRK$ python setup.py build
running build
running build_py
file snv_spectrum.py (for module snv_spectrum) not found
file snv_spectrum.py (for module snv_spectrum) not found

Some genomes exhibit very strong biases in their mutations between their reference and anti-reference bases. For example, human mitochondrial DNA shows a 4X higher mutation frequency for reference G->A mutations compared to reference C->T. It would be nice to either be able to plot out all 192 sequence contexts or, alternatively, have an option that converts to the same reference_notation via with_p[yrimidine, urine]_reference, but separately plot the spectra in a "reference" and "anti-reference" manner, as indicated in the attached figure.

Picture1.pdf

We provide a dynamic get_cosmic_signatures() helper function for pulling recent work from COSMIC.

Should we also support storage of signatures from other publications?

Would a JSON data store in a data/ folder be appropriate?