NodeSubstrates: Bridging Node-Link Diagrams with DR-Based Semantic Substrates

Abstract

Multivariate graph visualization faces a fundamental tension between representing network topology and showing attribute-driven structure. Existing techniques force analysts to choose between these perspectives, hindering the inspection of attribute-defined groupings within their structural context. Node-link diagrams convey connectivity but obscure attribute relationships, while dimensionality reduction (DR) projections expose attribute similarity at the cost of explicit edge information. We present NodeSubstrates, a hybrid visualization technique that integrates node-link diagrams with DR-based semantic substrates. In contrast to prior hybrid approaches such as NodeTrix, which replace subgraphs with topology-focused matrices, NodeSubstrates selectively repositions nodes using DR on their attributes, revealing clusters and outliers not visible in topological layouts. Substrate regions can be interactively created and dissolved. We demonstrate NodeSubstrates on a network of commercially available compounds, in which the hybrid representation supports drug design tasks spanning chemical attributes and graph structure. NodeSubstrates bridges graph drawing and dimensionality reduction by introducing a concrete hybrid technique for the emerging GD×DR research space.