Age-dependent DNA methylation Parenclitic Networks in family-based cohort patients with Down Syndrome

Network models are a powerful tool to represent, analyze and unfold the complexity of a large-dimensional data system at the fundamental level. The main advantage of network analysis is the opportunity to identify network disease signatures which we use in this paper for patients with Down Syndrome. One of the new methods based on the reconstruction of relations between system features is a Parenclitic Networks approach enabling setting links even for functionally unlinked features. In our work, we develop and generalize the Parenclitic Networks approach using Down Syndrome as a case study. We present our open-source implementation to make the method more accessible to all researchers. The software includes a complete workflow to construct Parenclitic Networks and demonstrate as a generalization that any machine learning algorithm can be chosen as a kernel to build edges in the network using geometric (SVM) and probabilistic (PDF) approaches as examples. We also present a new approach (PDF-adaptive) that allows to automatically to solve one of the main problems of building a network - choosing a “cut-off threshold”. We apply our implementation to the problem of detecting the network signature of Down Syndrome in DNA methylation data from the family-based cohort. We demonstrate the first insights into how Parenclitic Networks can be used not only as constructs to reduce data dimension and solve the classification problem using network characteristics, but also as a network signature of transitional changes.