Multilayer approach reveals organizational principles disrupted in breast cancer co-expression networks

The study of co-expression programs in the context of cancer can help to elucidate the genetic mechanisms that are altered and lead to the disease. The identification of gene co-expression patterns, unique to healthy profiles (and absent in cancer) is an important step in this direction. Networks are a good tool for achieving this as they allow to model local and global structural properties of the gene co-expression program. This is the case of gene co-expression networks (GCNs), where nodes or vertices represent genes and an edge between two nodes exists if the corresponding genes are co-expressed. Single threshold co-expression networks are often used for this purpose. However, important interactions in a broader co-expression space needed to unravel such mechanisms may be overlooked. In this work, we use a multilayer network approach that allows us to study co-expression as a discrete object, starting at weak levels of co-expression building itself upward towards the top co-expressing gene pairs.We use a multilayer GCNs (or simply GCNs), to compare healthy and breast cancer co-expression programs. By using the layers of the gene co-expression networks, we were able to identify a structural mechanism unique in the healthy GCN similar to well-known preferential attachment. We argue that this mechanism may be a reflection of an organizational principle that remains absent in the breast cancer co-expression program. By focusing on two well-defined set of nodes in the top co-expression layers of the GCNs—namely hubs and nodes in the main core of the network—we found a set of genes that is well conserved across the co-expression program. Specifically, we show that nodes with high inter-connectedness as opposed to high connectedness are conserved in the healthy GCN. This set of genes, we discuss, may partake in several different functional pathways in the regulatory program. Finally, we found that breast cancer GCN is composed of two different structural mechanisms, one that is random and is composed by most of the co-expression layers, and another non-random mechanism found only in the top co-expression layers.Overall, we are able to construct within this approach a portrait of the whole transcriptome co-expression program, thus providing a novel manner to study this complex biological phenomenon.