brioche-enrichment 0.1.2a0

Bayesian tests for set enrichment.

Bayesian tests for set enrichment.

Installation

Install from PyPI

pip install brioche-enrichment

or install from the Github repository

git clone [e-mail]:dpohanlon/brioche.git
pip install -r requirements.txt .

Usage

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:

$$n_{ij} = r_i + c_j + d_{ij}$$

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-")