Key points are not available for this paper at this time.
Introduction: Validating large knowledge graphs with the Shapes Constraint Language (SHACL) often yields violation reports too large to interpret and trace to root causes, especially in industry-scale datasets such as pharmaceutical omics pipelines. Methods: We present SHACLens, an interactive visualization workflow-developed with a major pharmaceutical partner-that links ontology, instance data, and violation reports across multiple coordinated views. We contribute a practitioner-informed workflow co-designed with pharmaceutical data-analysis experts. A Node-Link View combines ontology and groups of equivalent violations, a projection view reveals clusters of nodes with similar errors, a LineUp table combines instance data with violation information, a Class Tree offers a class-hierarchy overview, and an integrated LLM assistant provides contextual explanations and can operate the system via natural-language commands. Results: Within this workflow, selections and filters propagate across views, exposing co-occurring errors and their likely upstream causes. Analysts iteratively identify violation clusters, inspect correlations, and trace the detailed cause of errors. Evaluation and implications: We evaluated SHACLens through an iterative expert-in-the-loop design process with the partner team and a qualitative study on a transcriptomics dataset containing 5,203 violating nodes with the same experts. In this study, SHACLens efficiently surfaced repeated sets of errors due to missing objects and schema inconsistencies, supporting goal-oriented analysis and serendipitous findings.
Steinparz et al. (Mon,) studied this question.