Geranium: Multimodal Retrieval of Genomics Data Visualizations

Area of Science:

• Genomics
• Data Visualization
• Bioinformatics

Background:

• Interpreting genomics data requires effective visualization, but finding reusable examples is challenging.
• Current tools lack robust search capabilities for the extensive collection of genomics visualizations.
• This inefficiency hinders the creation of new visualizations, increasing authoring time.

Purpose of the Study:

• To introduce Geranium, a novel data visualization retrieval system designed for genomics.
• To enhance the search and authoring process for genomics visualizations.
• To provide researchers with adaptable visualization templates.

Main Methods:

• Geranium employs multimodal retrieval, accepting image, text, or grammar-based queries.
• It integrates grammar-based, multimodal, and text embeddings for comprehensive knowledge representation.
• Each visualization is represented multimodally, including Gosling specifications, renderings, and descriptions.

Main Results:

• The system supports multimodal retrieval for efficient searching of genomics visualizations.
• Retrieved examples act as templates, simplifying the adaptation and authoring of new visualizations.
• A collection of 3,200 visualizations across 50 categories is available, supporting diverse genomics applications.

Conclusions:

• Geranium significantly improves the efficiency and accessibility of genomics data visualization.
• The multimodal retrieval and template-based authoring streamline the visualization workflow for researchers.
• This system facilitates the exploration of complex genomics data, from single-cell epigenomics to structural variations.