Prior-informed conditional Gaussian graphical models for reconstructing personalised molecular networks, such as protein-protein interaction networks, from omics data.
prCGGM estimates a Gaussian graphical model whose mean and precision matrix depend linearly on subject-level covariates, Ω(X) = Δ₀ + Σₕ Xₕ Δₕ, and integrates prior knowledge on edges (for example ‘STRING’ confidence scores) through a weighted sparse-group lasso penalty applied only to the population-level network Δ₀. Covariate-specific changes Δₕ stay entirely data-driven. Personalised networks of new individuals are obtained from their covariates.
The method is described in:
Mapelli, A., Massi, M. C., Cuccuru, G., Di Angelantonio, E. and Ieva, F. (2026). Prior-informed conditional Gaussian graphical models: an application to protein interaction network reconstruction. arXiv:2606.31805.
install.packages("prCGGM")library(prCGGM)
data(prcggm_example)
set.seed(1)
fit <- prcggm(prcggm_example$x,
covariates = prcggm_example$covariates,
prior = prcggm_example$prior)
summary(fit)
plot(fit) # baseline network
plot(fit, component = "x1_1") # change associated with x1 = 1
# Personalised networks of new individuals
nets <- predict(fit, data.frame(x1 = factor(c("0", "1"))))
compare_networks(nets$Subject_1, nets$Subject_2, method = "frobenius")Depending on the inputs, prcggm() fits a standard GGM
(x only), a prior-weighted GGM (prior), a
conditional GGM (covariates) or the full prior-informed
conditional GGM (both). Node-wise regressions run sequentially, in
parallel (ncores), or as SLURM array jobs with
prcggm_slurm_prepare(), prcggm_node() and
prcggm_slurm_collect().
See vignette("prCGGM") for a complete walk-through.
The scripts used for the simulation study and the UK Biobank application are available at https://github.com/AlessiaMapelli/Prior-informed-conditional-GGMs.