Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GQVis: A Dataset of Genomics Data Questions and Visualizations for Generative AI

Skylar Sargent Walters, Arthea Valderrama, Thomas C. Smits, David Kouřil, Huyen N. Nguyen, Sehi L'Yi, Devin Lange, Nils Gehlenborg

TL;DR

The paper tackles the lack of domain-specific NL2VIS data for genomics by introducing GQVis, a large-scale dataset that links natural-language queries to Gosling-based visualizations, enriched with justifications, figure captions, and AltGosling text. It expands the DQVis pipeline to accommodate genomics data formats and domain conventions, enabling interactive multi-view visualizations across structural, functional, and epigenomic modalities. The authors present a five-stage generation process (template generation, expansion, multi-step curation, paraphrasing, quality review) and data sources from 4DN, ENCODE, and Chromoscope, yielding over 2.2 million data points (1.14M single queries, 628k query pairs, 589k chains). They also propose a quality-assessment framework and plan fine-tuning of domain-adapted NL2VIS models to improve genomic visualization authoring, accessibility, and multimodal search capabilities, thereby accelerating exploratory genomics analysis and interpretation.

Abstract

Data visualization is a fundamental tool in genomics research, enabling the exploration, interpretation, and communication of complex genomic features. While machine learning models show promise for transforming data into insightful visualizations, current models lack the training foundation for domain-specific tasks. In an effort to provide a foundational resource for genomics-focused model training, we present a framework for generating a dataset that pairs abstract, low-level questions about genomics data with corresponding visualizations. Building on prior work with statistical plots, our approach adapts to the complexity of genomics data and the specialized representations used to depict them. We further incorporate multiple linked queries and visualizations, along with justifications for design choices, figure captions, and image alt-texts for each item in the dataset. We use genomics data retrieved from three distinct genomics data repositories (4DN, ENCODE, Chromoscope) to produce GQVis: a dataset consisting of 1.14 million single-query data points, 628k query pairs, and 589k query chains. The GQVis dataset and generation code are available at https://huggingface.co/datasets/HIDIVE/GQVis and https://github.com/hms-dbmi/GQVis-Generation.

GQVis: A Dataset of Genomics Data Questions and Visualizations for Generative AI

TL;DR

The paper tackles the lack of domain-specific NL2VIS data for genomics by introducing GQVis, a large-scale dataset that links natural-language queries to Gosling-based visualizations, enriched with justifications, figure captions, and AltGosling text. It expands the DQVis pipeline to accommodate genomics data formats and domain conventions, enabling interactive multi-view visualizations across structural, functional, and epigenomic modalities. The authors present a five-stage generation process (template generation, expansion, multi-step curation, paraphrasing, quality review) and data sources from 4DN, ENCODE, and Chromoscope, yielding over 2.2 million data points (1.14M single queries, 628k query pairs, 589k chains). They also propose a quality-assessment framework and plan fine-tuning of domain-adapted NL2VIS models to improve genomic visualization authoring, accessibility, and multimodal search capabilities, thereby accelerating exploratory genomics analysis and interpretation.

Abstract

Data visualization is a fundamental tool in genomics research, enabling the exploration, interpretation, and communication of complex genomic features. While machine learning models show promise for transforming data into insightful visualizations, current models lack the training foundation for domain-specific tasks. In an effort to provide a foundational resource for genomics-focused model training, we present a framework for generating a dataset that pairs abstract, low-level questions about genomics data with corresponding visualizations. Building on prior work with statistical plots, our approach adapts to the complexity of genomics data and the specialized representations used to depict them. We further incorporate multiple linked queries and visualizations, along with justifications for design choices, figure captions, and image alt-texts for each item in the dataset. We use genomics data retrieved from three distinct genomics data repositories (4DN, ENCODE, Chromoscope) to produce GQVis: a dataset consisting of 1.14 million single-query data points, 628k query pairs, and 589k query chains. The GQVis dataset and generation code are available at https://huggingface.co/datasets/HIDIVE/GQVis and https://github.com/hms-dbmi/GQVis-Generation.
Paper Structure (10 sections, 4 figures)

This paper contains 10 sections, 4 figures.

Table of Contents

  1. Introduction
  2. Background & Related Work
  3. Dataset Generation
  4. Template Generation
  5. Template Expansion
  6. Multi-step Query Curation
  7. Paraphrasing
  8. Quality Review Software
  9. Dataset Results
  10. Conclusion and Future Work

Figures (4)

  • Figure 1: Multi-step generation pipeline. Chains are constructed from pairs of queries, in which the end query of a given pair corresponds to the start query of the next pair. The type of transition will determine how to update specs within the chain.
  • Figure 2: The review interface for assessing the quality of GQVis. Visuals and queries can be reviewed.
  • Figure 3: Relative frequency of query-visualization pairs by task taxonomy nusrat_tasks_2019 after subsampling.
  • Figure 4: Diversity of the dataset. GQVis covers structural, functional, and epigenomic data, including Chromoscope visuals.