Representation Learning for Networks in Biology and Medicine: Advancements, Challenges, and Opportunities

  • 2021-04-11 00:20:00
  • Michelle M. Li, Kexin Huang, Marinka Zitnik
  • 93

Abstract

With the remarkable success of representation learning in providing powerfulpredictions and data insights, we have witnessed a rapid expansion ofrepresentation learning techniques into modeling, analysis, and learning withnetworks. Biomedical networks are universal descriptors of systems ofinteracting elements, from protein interactions to disease networks, all theway to healthcare systems and scientific knowledge. In this review, we putforward an observation that long-standing principles of network biology andmedicine -- while often unspoken in machine learning research -- can providethe conceptual grounding for representation learning, explain its currentsuccesses and limitations, and inform future advances. We synthesize a spectrumof algorithmic approaches that, at their core, leverage topological features toembed networks into compact vector spaces. We also provide a taxonomy ofbiomedical areas that are likely to benefit most from algorithmic innovation.Representation learning techniques are becoming essential for identifyingcausal variants underlying complex traits, disentangling behaviors of singlecells and their impact on health, and diagnosing and treating diseases withsafe and effective medicines.

 

Quick Read (beta)

loading the full paper ...