Bayesian tests for set enrichment.
Install from PyPI
pip install brioche-enrichmentor install from the Github repository
git clone git@github.com:dpohanlon/brioche.git
pip install -r requirements.txt .Prepare some data in a contingency table format, with row and column set annotations
row_names = ["a", "b", "c", ...]
col_names = ["l", "m", "n", ...]
data = np.array([[30, 27, 10, ...], [28, 25, 11, ...], [31, 29, 15, ...], ...])
Import Brioche and create the multi-set enrichment object
from brioche.multisetEnrichment import MultisetEnrichment
enrichment = MultisetEnrichment(data, col_names, row_names, likelihood_type="sum")Optionally, specify whether there are total-sum constraints on the rows or columns
enrichment = MultisetEnrichment(data, row_names, col_names, row_constraint = True,
col_constraint = False, likelihood_type="sum")likelihood_type determines the form of the likelihood model for each cell, and also therefore what the 'null' model is. Both assume that each cell is a combination of a row and a column term, and a term that describes the deviation of the cell from the row and column terms:
When likelihood_type = 'sum', these are combined with a sum (the default), and when likelihood_type = 'prod' these are multiplied.
Run the Markov-chain Monte Carlo inference on the enrichment model, retrieve samples from the posterior, and return a Pandas DataFrame of enrichment z scores
samples = enrichment.runMCMC(num_samples=10000)
results = enrichment.getSummary(samples)Plot model parameters and enrichment posterior probability distributions
from brioche.plot import plotDeviations
plotDeviations(samples, threshold=2, x_labels=col_names, y_labels=row_names, name="test-")
