GRN-Stability Selection

Stability selection is a generally applicable subsampling-based framework for feature selection, allowing to bound the number of false positives if the exchangeability condition holds. Here we apply it in the context of gene regulatory networks: given a network-inference algorithm, we return the set of edges selected by stability selection.

Install

$ pip install grn-stability-selection

Usage

import giraffe

def giraffe_wrapper(expression, motif, ppi, regularization = 0):
    return giraffe.Giraffe(expression, motif, ppi, regularization = regularization).get_regulation()

stability = StabilitySelection(giraffe_wrapper, lambdas = np.linspace(lambda_min, lambda_max, grid_size), K = 100, v = 30)
stable_edges = stability(expression = expression, motif = motif, ppi = ppi)

The example above runs giraffe, a matrix factorization based algorithm for gene regulatory network inference, with different values of the regularization parmeter. After assigning a relevance score to each edge, it applies thresholding as is in the stability selection paper.

Appreciation

This implementation is inspired by the scikit-learn compatible implementation by Thomas Huijskens and colleagues. Shout-out for their great work and useful blog post!

Citation

@article{meinshausen2010stability,
  title={Stability selection},
  author={Meinshausen, Nicolai and B{\"u}hlmann, Peter},
  journal={Journal of the Royal Statistical Society: Series B (Statistical Methodology)},
  volume={72},
  number={4},
  pages={417--473},
  year={2010},
  publisher={Wiley Online Library}
}