Here we present an approach to infer and analyze gene regulation programs as shaped by different bioprocesses (contexts or layers) as given by the combined effects of gene-gene interactions, non-coding RNAs, copy number variants, DNA methylation profiles, 3D genome chromatin structure, etc. Our method relies on the implementation of a class of probabilistic graphical models over tensor-valued sampling spaces representing multi-omic experimental sources.
We will exemplify its use by looking at multi-omic regulatory networks in breast cancer. This approach has shed light into novel phenomena such as miRNA-gene interactions, epigenomic epistasis and others. The method may be further extended by incorporating additional experimental data sources by following conditional independence modeling on a Markov Random Field fashion.