Abstract
Prader-Willi Syndrome (PWS), a rare multi-system disorder characterized by insatiable appetite, growth abnormalities, and cognitive delay, results from genetic defects in a paternally expressed region of chromosome 15, q11.2-q13. This region contains several protein-coding genes and several genes encoding small nucleolar RNA (snoRNAs), including the SNORD116 gene cluster, but their exact role in PWS remains unclear. Since snoRNAs have wide-ranging effects on protein expression and proteins interact in a complex network, the genetic aberrations causing PWS are likely to cause far-reaching indirect effects on protein expression and activity. Here, we mapped PWS gene expression data onto a human protein-protein interaction (PPI) network and used graph learning techniques to 1) identify the most impacted proteins and 2) suggest novel disease mechanisms. We adapted GeneEMBED, a network-based method originally developed to model genetic variants associated with Alzheimer's Disease. Specifically, we integrated PWS or control expression data with the PPI network, calculated node embeddings, and identified proteins with large differences between PWS and control embeddings. These candidate proteins were subjected to functional enrichment analysis to discover altered biological processes in PWS. Candidate proteins were highly enriched for glycosylated proteins. Analysis of candidate glycosylation enzymes suggested abnormalities in mucin-type O-glycosylation, fucosylation, and glycosaminoglycan synthesis. Defects in these glycosylation pathways have been linked to several PWS phenotypes, including obesity, cognitive delay, and production of secondary sex hormones. Homeobox proteins, master regulators of transcription during development, were also overrepresented among the candidate proteins. In particular, we identified homeobox proteins that drive development of GABAergic and dopaminergic neurons. These neuronal pathways regulate appetite and other behaviors that are abnormal in individuals with PWS. Our results were highly reproducible across PWS model systems. This work offers new avenues for further research in PWS and provides a promising approach that can be applied to other complex diseases.