SHACLens: a visualization workflow for SHACL violation exploration in knowledge graphs.

Area of Science:

• Data Visualization
• Knowledge Graphs
• Bioinformatics

Background:

• Validating large knowledge graphs using Shapes Constraint Language (SHACL) generates extensive violation reports, hindering root cause analysis in industry-scale datasets, particularly in pharmaceutical omics.
• Interpreting and tracing violations in complex datasets like pharmaceutical omics pipelines is challenging due to the sheer volume of information.

Purpose of the Study:

• To introduce SHACLens, an interactive visualization workflow designed to address the challenges of interpreting large SHACL violation reports.
• To facilitate the identification and tracing of root causes for violations in large-scale knowledge graphs within the pharmaceutical industry.

Main Methods:

• Developed SHACLens, an interactive visualization workflow co-designed with pharmaceutical data analysis experts.
• Integrated multiple coordinated views including Node-Link View, projection view, LineUp table, and Class Tree.
• Incorporated an LLM assistant for contextual explanations and natural language control.

Main Results:

• SHACLens links ontology, instance data, and violation reports across coordinated views, enabling analysts to trace errors.
• Selections and filters propagate across views, exposing co-occurring errors and their likely upstream causes.
• The system efficiently surfaced repeated errors like missing objects and schema inconsistencies.

Conclusions:

• SHACLens provides an effective solution for navigating and understanding large SHACL violation reports in pharmaceutical omics.
• The interactive workflow supports both goal-oriented analysis and serendipitous discovery of data quality issues.
• Expert-in-the-loop design and qualitative studies confirm SHACLens's utility for data scientists and bioinformaticians.