Whole-Graph Embedding and Adversarial Attacks for Life Sciences

Networks provide a suitable model for many scientic and technological problemsthat require the representation of complex entities and their relations. Life sciencesapplications include systems biology, where molecular components are representedin integrated systems in which the interactions among them provide richer infor-mation than single components taken separately, or neuroimaging, where brainnetworks allow representing the connectivity between dierent brain locations. Inthe examples we focus on, a set of networks is available, with each network rep-resenting an entity (e.g., a molecule, a macro molecule, or a patient) and linksexpressing their relation in the chemical/biological domain.The growing size and complexity of biomedical networks and the high computa-tional complexity of graph analysis methods have lead to the investigation of theso-called whole-graph embedding techniques. Here, graphs are projected into lowerdimensional vector spaces, while retaining their structural properties, allowing toreducing the data complexity at the same time keeping the topological and struc-tural information. These techniques are showing very promising results in terms oftheir usability and potential. However, little research has focused on the analysisof their reliability and robustness. This need is strongly felt for real world ap-plications, where corrupted data, either due to acquisition noise or to intentionalattacks, could lead to misleading conclusions for the task at hand.Our objective here is to investigate on the adoption of adversarial attacks to whole-graph embedding methods for evaluating their robustness for classication in ap-plications of interest for life sciences.