This repository proposes a python implementation for stabilized ICA decomposition algorithm. Most of the technical details can be found in the references [1], [2],[3], [4] and [5].
Our algorithm integrates two well-known methods to solve the ICA problem :
- FastICA (implemented in scikit-learn)
- Preconditioned ICA for Real Data - Picard (implemented in picard package)
We also propose an implementation of the Mutual Nearest Neighbors method as well as a visualization tool to draw the associated network. It is used to study the stability of the ICA components through different datasets.
Stabilized-ica is now compatible with scikit-learn API, meaning that you can use the base class as a sklearn transformer and include it in complex ML pipelines.
pip install stabilized-ica
pip install git+https://github.com/ncaptier/stabilized-ica.git
We provide three jupyter notebooks for an illustration with transcriptomic data :
- ICA decomposition with stabilized ICA
- Stability of ICA components accross several NSCLC cohorts
- Stabilized ICA for single-cell expression data (cell cycle)
We provide one jupyter notebook for an illustration with EEG/MEG data :
We provide one jupyter notebook for an illustration of the integration of stabilized-ica into scikit-learn Machine learning pipelines:
The data set which goes with the jupyter notebook "ICA decomposition with stabilized ICA" can be found in the .zip file data.zip. Please extract locally the data set before running the notebook.
For the jupyter notebooks "Stability of ICA components accross several NSCLC cohorts" and "Stabilized ICA for single-cell expression data (cell cycle)" please note that you will have to load the data yourself in order to run them (all the necessary links are reported on the notebooks).
stabilized-ica was originally developped to deconvolute omics data into reproducible biological sources. We provide two additional computational tools to use stabilized-ica with omics data and interpret the extacted stable sources:
- sica-omics is a Python toolbox which complements stabilized-ica for the analysis of omics data. In particular, it proposes annotation functions to decipher the biological meaning of the extracted ica sources, as well as a wrapper to adapt stabilized-ica base code to the special case of Anndata format which is popular for dealing with single-cell gene expression data.
- BIODICA is a computational environment for application of independent component analysis (ICA) to bulk and single-cell molecular profiles, interpretation of the results in terms of biological functions and correlation with metadata. It uses the stabilized-ica package as its computational core. In particular, it comes with Graphical User interface providing a no-code access to all of its functionnalities.
import pandas as pd
from sica.base import StabilizedICA
df = pd.read_csv("data.csv", index_col=0)
sICA = StabilizedICA(n_components=45, n_runs=30 ,plot=True, n_jobs = -1)
sICA.fit(df)
Metagenes = pd.DataFrame(sICA.S_, columns = df.columns, index = ['metagene ' + str(i) for i in range(sICA.S_.shape[0])])
Metagenes.head()from sica.mutualknn import MNNgraph
cg = MNNgraph(data = [df1 , df2 , df3] , names=['dataframe1' , 'dataframe2' , 'dataframe3'] , k=1)
cg.draw(colors = ['r', 'g' , 'b'] , spacing = 2)
cg.export_json("example.json")If you use stabilized-ica in a scientific publication, we would appreciate citation to the following paper:
Nicolas Captier, Jane Merlevede, Askhat Molkenov, Ainur Seisenova, Altynbek Zhubanchaliyev, Petr V Nazarov, Emmanuel Barillot, Ulykbek Kairov, Andrei Zinovyev, BIODICA: a computational environment for Independent Component Analysis of omics data, Bioinformatics, Volume 38, Issue 10, 15 May 2022, Pages 2963–2964, https://doi.org/10.1093/bioinformatics/btac204
This package was created as a part of the PhD project of Nicolas Captier in the Computational Systems Biology of Cancer group of Institut Curie.
[1] "Determining the optimal number of independent components for reproducible transcriptomic data analysis" - Kairov et
al. 2017
[2] "Assessing reproducibility of matrix factorization methods in independent transcriptomes" - Cantini et al. 2019
[3] "Icasso: software for investigating the reliability of ICA estimates by clustering and visualization" - Himberg et
al. 2003
[4] "Faster independent component analysis by preconditioning with Hessian approximations" - Ablin et al. 2018
[5] "BIODICA: a computational environment for Independent Component Analysis of omics data" - Captier et al. 2022