robustica

Fully customizable robust Independent Component Analysis (ICA).

Description

This package contains 3 modules:

• RobustICA

  Defines the most important class that allows to perform and customize robust independent component analysis.

• InferComponents

  Retrieves the number of components that explain a user-defined percentage of variance.

• examples

  Contains handy functions to quickly create or access example datasets.

A more user-friendly documentation can be found at https://crg-cnag.github.io/robustica/.

Requirements

In brackets, versions of packages used to develop robustica.

• numpy (1.19.2)
• pandas (1.1.2)
• scipy (1.6.2)
• scikit-learn (0.23.2)
• joblib (1.0.1)
• tqdm (4.59.0)
• (optional) scikit-learn-extra (0.2.0): required only for clustering algorithms KMedoids and CommonNNClustering

Installation

[optional] scikit-learn-extra incompatibility

To use the clustering algorithms KMedoids and CommonNNClustering, install a forked version first to avoid incompatibility with the newest numpy (see #6 for more info on this).

pip install git+https://github.com/TimotheeMathieu/scikit-learn-extra

pip

pip install robustica

local (latest version)

git clone https://github.com/CRG-CNAG/robustica
cd robustica
pip install -e .

Usage

from robustica import RobustICA
from robustica.examples import make_sampledata

X = make_sampledata(ncol=300, nrow=2000, seed=123)

rica = RobustICA(n_components=10)
# note that by default, we use DBSCAN algorithm and the number of components can be smaller
# than the number of components defined.
S, A = rica.fit_transform(X)

# source matrix (nrow x n_components)
print(S.shape)
print(S)

(2000, 3) 
[[ 0.00975714  0.00619138  0.00502649]
 [-0.0021527  -0.0376857   0.0117938 ]
 [ 0.00046302  0.01712561  0.00518039]
 ...
 [ 0.00128344 -0.00767099  0.0047334 ]
 [ 0.00644422 -0.00498327  0.01325542]
 [ 0.0017873  -0.01739889 -0.00445954]]

# mixing matrix (ncol x n_components)
print(A.shape)
print(A)

(300, 3)
[[-1.79503194e-02 -1.05611924e+00  5.36688700e-01]
 [ 1.03342514e-01  7.43471382e-02  4.90472157e-01]
 [ 4.89753256e-01 -1.11300905e+00 -7.55809647e-01]
 ...
 [ 4.30468472e-01 -4.87992838e-01 -7.77965512e-01]
 [ 3.44078031e-02  4.09029805e-01 -7.29076312e-01]
 [ 2.15557427e-02  2.89301273e-01 -2.96690459e-01]]

Tutorials

• Basic pipeline for exploratory analysis
• Using a custom clustering class
• Inferring the number of components

Contact

This project has been fully developed at the Centre for Genomic Regulation within the group of Design of Biological Systems

Please, report any issues that you experience through this repository's "Issues" or email:

• Miquel Anglada-Girotto
• Sarah A. Head
• Luis Serrano

License

robustica is distributed under a BSD 3-Clause License (see LICENSE).

Citation

Anglada-Girotto, M., Miravet-Verde, S., Serrano, L., Head, S. A.. "robustica: customizable robust independent component analysis". BMC Bioinformatics 23, 519 (2022). DOI: https://doi.org/10.1186/s12859-022-05043-9

References

• Himberg, J., & Hyvarinen, A. "Icasso: software for investigating the reliability of ICA estimates by clustering and visualization". IEEE XIII Workshop on Neural Networks for Signal Processing (2003). DOI: https://doi.org/10.1109/NNSP.2003.1318025
• Sastry, Anand V., et al. "The Escherichia coli transcriptome mostly consists of independently regulated modules." Nature communications 10.1 (2019): 1-14. DOI: https://doi.org/10.1038/s41467-019-13483-w
• Kairov, U., Cantini, L., Greco, A. et al. Determining the optimal number of independent components for reproducible transcriptomic data analysis. BMC Genomics 18, 712 (2017). DOI: https://doi.org/10.1186/s12864-017-4112-9