Netpro2vec: A Graph Embedding Framework for Biomedical Applications

The ever-increasing importance of structured data in different applications, especially in the biomedical field, has driven the need for reducing its complexity through projections into a more manageable space. The latest methods for learning features on graphs focus mainly on the neighborhood of nodes and edges. Methods capable of providing a representation that looks beyond the single node neighborhood are kernel graphs. However, they produce handcrafted features unaccustomed with a generalized model. To reduce this gap, in this work we propose a neural embedding framework, based on probability distribution representations of graphs, named Netpro2vec. The goal is to look at basic node descriptions other than the degree, such as those induced by the Transition Matrix and Node Distance Distribution. Netpro2vec provides embeddings completely independent from the task and nature of the data. The framework is evaluated on synthetic and various real biomedical network datasets through a comprehensive experimental classification phase and is compared to well-known competitors.

1 Introduction

The constant growth and availability of biomedical data and the application of dedicated computational and mathematical techniques allow an accurate overview and simulation of the physio-pathological processes. In this scenario, a huge contribution has been provided by the advent of omics science (e.g., genomics, transcriptomics, proteomics, metabolomics), whose main objective is to qualitatively and quantitatively characterize the biological molecules (genes, proteins, metabolites)involved in the structure, function, and dynamics of a cell, tissue, or even a whole organism. Understanding the role of single components in a biological system is definitely crucial, but, to give an exhaustive interpretation of the underlying mechanisms and functioning, the importance of the relationships between these components cannot be ignored. The holistic view of biological molecule interactions is the main issue of systems biology, where a system is organized as a network structure (a.k.a. graph)and is mainly defined by the connections (edges)among its components (nodes). Based on the source data, involved molecules, and the objective of the study, different biomedical networks can be created, containing one type of entity or including various sources of data, integrated “horizontally” on the same layer or “vertically” to include different interaction layers.