Designing for Science at Scale
Cytoexplorer is a specialized application within the Polus platform — a comprehensive scientific data management ecosystem conceived as "Adobe Cloud for scientific research." Polus provides researchers with an integrated suite of interoperable tools spanning data collection, visualization, and publication-ready output.
Within that ecosystem, Cytoexplorer addresses one of the hardest problems in life sciences tooling: making high-content screening (HCS) and cytometry datasets — which can contain hundreds of thousands of individual cell data points — explorable and interpretable without requiring deep computational expertise. The application had grown organically alongside its engineering team, resulting in scattered controls, opaque terminology, and workflows that demanded navigating disconnected surfaces.
"A lot of this has to have a description — it's not self-evident."
Leading the UX Work
UX/UI Design Lead
I oversaw design direction and output across a 10-month sprint cycle, coordinating between two designers (Gurveen Sekhon, Sasa Chen), a UX researcher (Vanessa Zhou), the Lead Full Stack Software Engineer, and scientific stakeholders including the Associate Director of AI and the VP of Data Science. I facilitated weekly check-ins, set design priorities, and served as the primary design voice in stakeholder reviews. Tools: Adobe XD, Adobe Illustrator, Lucidchart, ChatGPT, Jira, Confluence.
What I Directly Designed
- Data Manager nav & toolbar consolidation — reorganized scattered controls into three logical sidebar groups
- Image Render module — defined layout approaches and drove stakeholder discussion on interaction model
- Charts/Tree split-screen — pop-out view enabling simultaneous chart and node navigation
- Polus suite shell alignment — updated header nav to align with the broader platform
- Clustering & dimension reduction UX — facilitated technical discovery then scoped design requirements
What I Directed & Oversaw
- Feature Manager interactions — directed Gurveen's iterative prototype work across 8+ sprint cycles
- Doc Explorer & LLM UI — guided design of document loading, filtering, and AI-chat comparison
- Accessibility & color system — conducted WCAG audit and developed the four-mode palette
- User personas & user flows — developed personas and directed user flow diagramming
- Competitive analysis — directed Vanessa's benchmarking research across scientific visualization tools
- Heuristic evaluation — led annotated analysis of the legacy interface
Building a Research Foundation
Before any redesign work began, the team established a rigorous research foundation across five methodologies: heuristic evaluation, user personas, user flow diagramming, accessibility checks, and competitive analysis. My role was to direct the program and ensure findings fed directly into design decisions throughout the sprint cycle.
Heuristic Evaluation
We conducted a systematic evaluation of the existing interface, annotating usability violations across the legacy UI. Key findings included information overload in the feature column list, unlabeled icons throughout the toolbar, ambiguous CTAs, and poor accessibility — particularly contrast ratios and lack of chart separation — across the Chart Explorer and Class Explorer views.
User Personas
Five researcher personas were developed to ground design decisions in the actual user base — spanning research scientists, lab directors, computational biologists, and bioengineers. These anchored design priorities across all sprints and were referenced whenever stakeholder opinions diverged on feature scope.
User Flow Diagrams
We mapped primary workflows through Cytoexplorer using Lucidchart as shared alignment artifacts for engineering and science partners. As the platform expanded to include Doc Explorer and LLM-powered querying, we created a dedicated AI Document Analysis flow — mapping the full path from collection selection through LLM query, interactive article exploration, dimension reduction, and save/export.
Accessibility Check
A WCAG-based accessibility audit evaluated contrast ratios, interactive element sizing, keyboard navigability, and screen reader compatibility across light and dark mode variants. Findings directly informed the four-mode color system and the component library's interactive states.
Competitive Analysis
Vanessa benchmarked Cytoexplorer against comparable scientific visualization and data exploration tools, identifying industry conventions and patterns the team could adopt or deliberately depart from to better serve the research context.
From Legacy UI to Cohesive System
The Starting Point — Legacy Interface
The original Cytoexplorer interface was built developer-first — functional but visually dense, with controls distributed across disconnected panel areas, no tooltip system, and inconsistent interaction patterns across modules.
Visual Design System
Working within Polus design standards, I oversaw the development of a structured visual system covering a four-mode color palette (Light/Dark × High Contrast/Regular Contrast), a comprehensive icon library in both light and dark variants, and a Roboto-based type scale from H1 through Body 2 — ensuring consistency and accessibility compliance across all modules and environments.
Component Library
A comprehensive component library was built covering buttons, tabs, pagination, nav icons, and the Doc Explorer data panel — all specified across four color modes with default, hover, selected, and inactive states. This gave the distributed team a shared vocabulary and ensured interaction consistency as different designers worked on different modules simultaneously.
Feature Manager
The Feature Manager — responsible for selecting, grouping, and filtering hundreds of data columns — underwent the most intensive iterative design work of the project. Gurveen and I worked through 8+ prototype cycles, resolving questions around group creation flows, edit states, filter controls, and the Selected Columns panel. The redesign introduced structured Groups/Dive/Facet/Processing tabs, a percentage-based filter slider, and a dedicated group creation modal.
Data Manager
I led the consolidation of the Data Manager's toolbar — previously scattered across the top nav — into a structured left sidebar with three logical groups: selection tools, behaviors, and display options. The Charts/Tree split panel enabled researchers to navigate node trees and charts simultaneously. A contextual Behaviors popover gave users access to multigroup selection, subset selection, axis flip, and binning resolution — all accessible without leaving the main view.
Doc Explorer & LLM Integration
Late in the project, the VP of Data Science requested a Doc Explorer module — a document-centric exploration interface with LLM-powered querying and summarization. Gurveen designed the interface across several sprint cycles, resolving key questions around query scope, model selection, document summarization with save-to-doc capability, and how to display article content alongside the UMAP plot context.
Interactive Prototypes
All design work was presented as clickable prototypes — not static mockups — every sprint. This kept stakeholder feedback grounded in actual interaction, surfaced edge cases early, and ensured engineering effort was never committed to unvalidated assumptions.
Ten Months of Sprint Work
The project ran as a continuous sprint cycle from March 2024 through January 2025, combining weekly UX prototype presentations with open technical discussions with engineering and science stakeholders. Notable dates: kickoff March 1, 2024 · stakeholder approval review October 15, 2024 · project suspension January 14, 2025.
"Prototypes are conversations, not deliverables."
Discovery & Heuristic Review
Live walkthrough of the existing interface. Identified missing tooltips, scattered controls, and unlabeled icons. Kicked off the formal UX research program.
UX Research Program
Directed five personas, user flow diagrams, heuristic evaluation, accessibility audit, and competitive analysis. Established a prioritized use case list with scientific partners.
Feature Manager & Navigation Redesign
Led iterative design of the Feature Manager grouping flow across multiple prototype reviews. Redesigned the toolbar into a structured sidebar and aligned the shell with the Polus suite.
Image Render Module
Designed the UX for rendering cell images linked to scatter plot data points. Explored two structural approaches and resolved critical interaction questions with stakeholders.
Clustering, Dimension Reduction & Explorer Expansion
Organized a technical briefing on clustering and UMAP/PCA-style reduction, translating the science into UX requirements. Added Explorer as a top-level nav section.
Stakeholder Approval Review
Presented the full redesigned system to the VP of Data Science and the broader team. Used as the formal approval checkpoint before moving into Doc Explorer work.
Doc Explorer & LLM Integration
Directed design of document loading, LLM query interface, article summarization, and document comparison with inline AI output and source linking.
Final Refinements & Suspension
Delivered the full-screen Chat with Data modal, comparison view, and save results flow. User testing was scheduled — then NIH/DOGE budget cuts in January 2025 led to the offboarding of the Axle UX team, and work on the project stopped.
Before & After
The comparisons below show the transformation across all three core modules — from the original developer-built interface to the redesigned system.
File Manager
Feature Manager
Data Manager
Outcomes
Sprint cycles of iterative prototype delivery
Major modules designed or redesigned
UX research methodologies applied
Project status: As user testing with scientists in molecular modeling, biology/engineering, and AI was being scheduled, NIH/DOGE budget cuts in January 2025 led to the offboarding of the Axle Informatics UX team — one of several Polus suite projects brought to a halt as a result. Despite this, Cytoexplorer achieved substantial improvements in usability and produced a well-documented design system and prototype library for any future continuation.
Key Takeaways
- Complexity demands progressive disclosure. When users are navigating datasets with hundreds of thousands of data points, surfacing everything at once isn't power — it's noise. The design work here was fundamentally about deciding what to show, when, and why.
- Terminology is part of the interface. Working alongside bioinformaticians and cell biologists made it clear that a wrong label isn't just confusing — it undermines trust. Every term was validated against how scientists actually spoke about their work.
- Cross-functional collaboration sharpens decisions. Designing alongside engineers and domain experts forced every design choice to be defensible. That pressure produced a tighter, more considered system than solo work would have.
- Good design work outlasts the project. Even with Cytoexplorer halted, the documented design system, component library, and prototype archive remain a handoff-ready foundation. Work done well doesn't disappear when a project does.