euxhenh / phenotype-cover

A package for biomarker selection based on multiset multicover and the cross-entropy-method.

This repository provides two algorithms for the phenotype cover (PC) biomarker selection problem introduced in the paper: "Multiset multicover methods for discriminative marker selection". GreedyPC is based on the extended greedy algorithm to set cover, and CEM-PC is based on the cross-entropy-method.

Install via

pip install multiset-multicover
pip install phenotype-cover

Other packages that phenotype-cover depends on are numpy, matplotlib, and scikit-learn.

Import GreedyPC or CEMPC from phenotype_cover.

Example

>>> from phenotype_cover import GreedyPC
>>> from sklearn.datasets import make_classification

>>> # You may need to log-transform X if working with raw counts
>>> X, y = make_classification(1000, 200, n_informative=5, n_classes=5, scale=100)

>>> gpc = GreedyPC()
>>> gpc.fit(X, y)
>>> features = gpc.select(100)  # coverage of 100

Some other functionality implemented in GreedyPC

>>> # Number of elements reamining and coverage attained after every iteration
>>> gpc.plot_progress()
>>> gpc.n_elements_remaining_per_iter_
>>> gpc.coverage_per_iter_

>>> # Heatmap of the coverage provided by some feature i
>>> gpc.feature_coverage(i)

>>> # Maximum possible coverage for evey class pair
>>> gpc.max_coverage()

>>> # Pairs that could not be covered to the desired `coverage`
>>> gpc.pairs_with_incomplete_cover_

phenotype-cover's Issues

ValueError in example

I tried using your method but running the example code provided in the readme,

X = np.log1p(np.random.randint(0, 1000, (1000, 200)))
y = np.random.randint(0, 5, 1000)
gpc = GreedyPC()
gpc.fit(X, y)
features = gpc.select(100)

produces the following error:

Traceback (most recent call last):
  File ".../gpc_test.py", line 17, in <module>
    features = gpc.select(100)
  File ".../.local/lib/python3.10/site-packages/phenotype_cover/_phenotype_cover.py", line 169, in select
    self.multiplier = self._multiplier
  File ".../.local/lib/python3.10/site-packages/phenotype_cover/_phenotype_cover.py", line 291, in multiplier
    self._gci_wrapper.fit(self.M_)
  File ".../.local/lib/python3.10/site-packages/phenotype_cover/_gci_wrapper.py", line 98, in fit
    raise ValueError(
ValueError: No sets were selected. This means your data matrix consists of only zeros. Please use higher multiplicity.

Python version 3.10 on Ubuntu Ubuntu 22.04.2 LTS.

Could You please help me with this issue?

Thank You in advance,
Bartol