Context-Driven Visualization Workflows for Biology and Medicine

Date of Award

Fall 1-1-2025

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Computer Science

First Advisor

Blenman, Kim

Abstract

Visualizations are a crucial part of omics, both in analysis and subsequent communication of insights. There is a plethora of visualization types in omics, but a lack of emphasis on a structured method of integrating insights across multiple visualizations. Furthermore, the visualizations are difficult to interpret even for domain experts due to the lack of biological context in the visualizations. The work in this dissertation defines a new problem space that applies the concept of guidance and narrative visualization from visualization research in novel ways to omics visualization. We call this problem space "context-driven visualization workflows" for the fields of biology and medicine. We explore how to build guidance into visualization workflows for biologists to streamline insight discovery and construct narrative visualizations that communicate their insights using biological context. The resulting narrative visualizations are understood by experts and non-experts alike. We took an iterative approach in which we included domain experts and end user biologists during each phase of development. We first established an appropriate form of guidance leading users to view omics visualizations in a coarse-to-fine manner, allowing them to perceive overall trends in the data and then narrow down to sections of interest. While obvious, the coarse-to-fine approach is not a standardized method of analysis in omics. We conducted a user study with domain experts and verified that the coarse-to-fine workflow is preferred over unstructured workflows common to omics. While many types of omics visualizations exist, they typically do not include biological context, which we consider to be depictions of associations with specific anatomical and cellular components of the human body (e.g., organs (liver), organelles (mitochondria), cells (B cells)). Since the same biological entities can exist in multiple biological processes, without depiction of context it is difficult to interpret omics visualizations, possibly leading to false scientific conclusions. To address this gap, we developed the context map visualization type. Our context map interface allows users to build a narrative visualization communicating their biological insights with visual context. Our interface adopts a semi-automated approach, allowing users to combine their own expertise with information from existing biological pathways to produce novel insights. We evaluated our context map interface with end user biologists and determined that biological pathway visualizations containing biological context are preferred over the domain standard manual generation of pathway maps using tools such as PowerPoint. Apart from biologists, omics visualizations also need to be understood by those without domain expertise. Omics research teams often consist of people from non-biology and non-medical backgrounds, so visualizations are most effective when they can communicate to both domain and non-domain experts alike. We conducted a two-phase study to evaluate the comprehensibility of the context map among non-domain experts, and concluded that the context map visualization type communicated domain insight more accurately compared to the existing pathway map standard to omics. Lastly, in our interdisciplinary work we discovered a critical accessibility gap between computer science research prototypes and their target end users. Oftentimes, research prototypes are not set up for wide-spread distribution and are not maintained after the work is published. However, end users do not always have the expertise to debug code and technical challenges that arise during installation and runtime. The accessibility issue is exacerbated by the numerous dependencies incorporated into each library in software development, and lack of maintenance across dependencies. This eventually leads to deterioration of software performance, or worse not being able to start up at all. This accessibility gap prevents many end users from benefitting from computer science research advances, limiting progress in their own fields. Furthermore, it also prevents end users from giving valuable feedback on the software that would help further computer science research. Our prototype context-driven visualization workflow, visAPPprot, demonstrates how to incorporate domain context into visualizations, guide narrative visualization for domain experts, and streamline visualization software distribution for generalist users to make data visualization more accessible.

This document is currently not available here.

Share

COinS