coglasso - Collaborative Graphical Lasso

Coglasso implements collaborative graphical lasso, an algorithm for network reconstruction from multi-omics data sets (Albanese, Kohlen and Behrouzi, 2024). Our algorithm joins the principles of the graphical lasso by Friedman, Hastie and Tibshirani (2008) and collaborative regression by Gross and Tibshirani (2015).

Installing coglasso

You can install the CRAN release of coglasso with:

install.packages("coglasso")

Installing the development version

To install the development version of coglasso from GitHub you need to make sure to install devtools with:

if (!require("devtools")) {
  install.packages("devtools")
}

You can then install the development version with:

devtools::install_github("DrQuestion/coglasso")

Usage

Here follows an example on how to reconstruct and select a multi-omics network with collaborative graphical lasso. For a more exhaustive example we refer the user to the vignette vignette("coglasso"). The package provides example multi-omics data sets of different dimensions, here we will use multi_omics_sd_small. The current version of the coglasso package accepts multi-omics data sets with multiple “omic” layers, where the single layers are grouped by column. For example, in multi_omics_sd_small the first 14 columns represent transcript abundances, and the other 5 columns represent metabolite abundances. The function to perform both network estimation and network selection is bs(). The suggested usage of bs() only needs the input data set, the dimensions of the “omic” layers, and the number of values to explore for each hyperparameter.

library(coglasso)

sel_cg <- bs(multi_omics_sd_small, pX = c(14, 5), nlambda_w = 15, nlambda_b = 15, nc = 5)

# To see information about the network estimation and selection
print(sel_cg)

bs() explores several combinations of the hyperparameters characterizing collaborative graphical lasso. Then, it selects the combination yielding the best network according to the chosen model selection method. Among others, this function implements eXtended Efficient StARS (XEStARS), a significantly faster and memory-efficient version of eXtended StARS (XStARS, Albanese, Kohlen and Behrouzi, 2024). These are coglasso-adapted versions of the StARS selection algorithm (Liu, Roeder and Wasserman, 2010) selecting the hyperparameter combination that yields the most stable, yet sparse network. XEStARS is the default option for the parameter method, so it is enough to enjoy the comfort of the default behaviour and let the function do the rest. To plot the selected network, use:

plot(sel_cg)

References

Albanese, A., Kohlen, W., & Behrouzi, P. (2024). Collaborative graphical lasso (arXiv:2403.18602). arXiv https://doi.org/10.48550/arXiv.2403.18602

Friedman, J., Hastie, T., & Tibshirani, R. (2008). Sparse inverse covariance estimation with the graphical lasso. Biostatistics, 9(3), 432–441. https://doi.org/10.1093/biostatistics/kxm045

Gross, S. M., & Tibshirani, R. (2015). Collaborative regression. Biostatistics, 16(2), 326–338. https://doi.org/10.1093/biostatistics/kxu047

Liu, H., Roeder, K., & Wasserman, L. (2010). Stability Approach to Regularization Selection (StARS) for High Dimensional Graphical Models (arXiv:1006.3316). arXiv https://doi.org/10.48550/arXiv.1006.3316