GeoMx Data Analysis Setup

Workflow for researchers preparing genomic sequencing data for spatial analysis. This would become a feature of the ROSALIND bioinformatics platform, intended to be used alongside image workflows and to bring more biopharma teams onto platform.

Workflow for researchers preparing genomic sequencing data for spatial analysis. This would become a feature of the ROSALIND bioinformatics platform, intended to be used alongside image workflows and to bring more biopharma teams onto platform.

Role

Product Designer

Timeline

6 months

TEam Context

Full Stack Developer, Bioinformatic Scientist, Head of Innovation, Product Manager, Customer Success Manager

PLATFORM

Web App

Rosalind GeoMx quality-control screen with a table of AOI samples beside a gene detection histogram that has a highlighted cutoff range and a slider.

Overview

Overview

Figure in a lab coat with a hand on their face, looking downtrodden at a laptop with a speech bubble of code.
the Problem
the Problem
  • Spatial analysis meant writing scripts or using software that made results hard to share.


  • Analysis method choices could skew results and lead to false postives.

  • Spatial analysis meant writing scripts or using software that made results hard to share.


  • Analysis method choices could skew results and lead to false postives.

Simplified illustrations of Rosalind dashboards, a lightbulb flashing above them.
the business Need
the business Need
  • A workflow without the need for coding expertise, starting from raw data files.


  • Make GeoMx experiments fit seamlessly with the platform's sharing and meta-analysis capabilites.

  • A workflow without the need for coding expertise, starting from raw data files.


  • Make GeoMx experiments fit seamlessly with the platform's sharing and meta-analysis capabilites.

A set of jigsaw puzzle pieces fitting together, the one in the middle is highlighted.
my role
my role
  • Designed the final workflow from end-to-end.


  • Owned the error and warning states that bring clarity to issues with data quality.


  • Partnered with engineering to apply the design system's data viz components.

  • Designed the final workflow from end-to-end.


  • Owned the error and warning states that bring clarity to issues with data quality.


  • Partnered with engineering to apply the design system's data viz components.

Outcomes

A guided workflow that lets users adjust QC parameters in real time and makes the results visible.

A guided workflow that lets users adjust QC parameters in real time and makes the results visible.

8 updated data visualization and system alert components, with documentation.

8 updated data visualization and system alert components, with documentation.

A web-based tool that brought on 25+ new users/teams to the platform.

A web-based tool that brought on 25+ new users/teams to the platform.

8

updated components

25+

New users

52 hours

Saved on design debt fixes

Research & Strategy

Research & Strategy

Spatial profiling hardware had outpaced the analysis tools. The researchers who owned the data couldn’t reliably see or undo analysis decisions (QC, normalization method, thresholds) that affected their results.

Spatial profiling hardware had outpaced the analysis tools. The researchers who owned the data couldn’t reliably see or undo analysis decisions (QC, normalization method, thresholds) that affected their results.

Frustrated researcher in a lab coat, palm pressed to forehead and slumped shoulders, in front of a monitor with a code terminal, cluttered excel sheet, and uneven box graphs

Key Pain Points

Key Pain Points

"I can't tell what's a real data signal or what's just noise."

"I can't tell what's a real data signal or what's just noise."

"I can't tell what's a real data signal or what's just noise."

Samples with different signal intensities can lead to poor quality data and biased analyses.

Samples with different signal intensities can lead to poor quality data and biased analyses.

Source: van Hijfte et al., iScience

"I don't know which settings are right for my data."

"I don't know which settings are right for my data."

Normalization and statistical model choices change results, and the defaults aren’t validated for every dataset.

Normalization and statistical model choices change results, and the defaults aren’t validated for every dataset.

Source: van Hijfte et al., iScience

"I need to code, or find a bioinformatician, to analyze my own data."

"I need to code, or find a bioinformatician, to analyze my own data."

Analysis meant going into R/Python or the software built into the spatial instrument.

Analysis meant going into R/Python or the software built into the spatial instrument.

Source: van Hijfte et al., iScience

"Cleaning my data feels risky and I can't take it back."

"Cleaning my data feels risky and I can't take it back."

The wrong applied threshold could drop key samples or genes, and users couldn’t see what they’d lose.

The wrong applied threshold could drop key samples or genes, and users couldn’t see what they’d lose.

Source: van Hijfte et al., iScience

The Strategy

The solution needed to extend to what the platform already does well, which is guided, no-code workflows in the cloud. My role was to keep the design consistent with updated components, keep data visual and interactive, and account for various error/alert states.

The solution needed to extend to what the platform already does well, which is guided, no-code workflows in the cloud. My role was to keep the design consistent with updated components, keep data visual and interactive, and account for various error/alert states.

User Journey

Figure in a lab coat pointing at a monitor screen that has an upload icon next to a check symbol.

Data Import

Data Import

Upload raw DCC files and confirm the panel.

Upload raw DCC files and confirm the panel.

Figure in a lab coat with a lightbulb turning on over their head, looking at a floating table of data with the first two rows featuring an alert icon in the first column. The figure has a surprised expression.

Quality Check

Quality Check

Review sample QC, flag problems, and remove low quality data.

Review sample QC, flag problems, and remove low quality data.

Figure in a lab coat next to a large histogram, adjusting a slider control underneath it.

Adjust Parameters

Adjust Parameters

Reprocess data in real time by changing AOI and gene detection thresholds.

Reprocess data in real time by changing AOI and gene detection thresholds.

A researcher in a lab coat with a hand on their chin, looking at a Venn diagram in front of them.

Analyze

Analyze

Normalize, compare data subsets, and explore pathways.

Normalize, compare data subsets, and explore pathways.

Quick Wins

Quick Wins

Since the platform already built a guided workflow framework in other features, the first steps in the user journey only needed a few small adjustments.

Since the platform already built a guided workflow framework in other features, the first steps in the user journey only needed a few small adjustments.

Status indicators show the progress of the data upload.

Status indicators show the progress of the data upload.

An expandable list shows each file's upload status, so problems are easy to spot.

An expandable list shows each file's upload status, so problems are easy to spot.

Updated button components, including changing the "back" button to a back arrow icon.

Updated button components, including changing the "back" button to a back arrow icon.

Improved color contrast, following WCAG 2.0 recommendations.

Improved color contrast, following WCAG 2.0 recommendations.

Updated button components, including changing the "back" button to a back arrow icon.

Improved color contrast, following WCAG 2.0 recommendations.

Import Data screen with a progress bar and a list of DCC files, each with a status icon.
Import Data screen reading “2 of 14 files received and processing”, with a red error that minimum sample requirements aren’t met, a yellow warning about extra lab worksheets with Accept and Remove buttons, and an expandable list of per-file upload progress.

Designing for Error

Making the status of each file upload visible surfaces problems with the files right away and suggests paths for how to move forward.

Designing for Error

Making the status of each file upload visible surfaces problems with the files right away and suggests paths for how to move forward.

Designing for Error

Making the status of each file upload visible surfaces problems with the files right away and suggests paths for how to move forward.

Going Further

Going Further

The bigger design work was in the QC and normalization steps of the workflow. Making visual data clear at a glance was the main challenge of this project. New design components were created in parallel.

The bigger design work was in the QC and normalization steps of the workflow. Making visual data clear at a glance was the main challenge of this project. New design components were created in parallel.

QC screen for AOI sequencing performance, with a sample table of pass/fail flags beside a stacked bar chart of QC results.

Researchers can judge dataset quality at a glance by showing which AOIs pass or fail and suggesting which to remove.

Researchers can judge dataset quality at a glance by showing which AOIs pass or fail and suggesting which to remove.

QC screen for removing AOIs with low gene detection, with a sample table, a histogram with a highlighted cutoff range, and a threshold slider.

Data becomes interactive as users set a gene detection cutoff and see right away which AOIs fall below it.

Data becomes interactive as users set a gene detection cutoff and see right away which AOIs fall below it.

QC table where Samples 001–003 are marked FAIL in red and checked for removal, and Sample 004 is marked CAUTION in orange and unchecked.

Poor QC surfaced to the top, which may have otherwise been missed by the researcher.

Poor QC surfaced to the top, which may have otherwise been missed by the researcher.

Users find it easier to correct errors than make the full call themselves, which is why the system automatically marks failed samples for removal.

Users find it easier to correct errors than make the full call themselves, which is why the system automatically marks failed samples for removal.

Change Parameters dialog listing QC thresholds such as Minimum AOI Reads and Minimum % Stitched, with values changed from the default highlighted in blue.

The system automatically suggests parameters to the user AND allows them to make adjustments. Changes to the default are visually communicated through the blue highlighting.

The system automatically suggests parameters to the user AND allows them to make adjustments. Changes to the default are visually communicated through the blue highlighting.

Early normalization screen explaining processing methods in a text panel.
Later normalization iteration that adds a small chart comparing methods, with the method dropdown open.

The first iterations of the normalization step attempted to explain the different processing methods through text and then data visualization. Making sure that the user understands the impact this step has on their data is significant for downstream.

The first iterations of the normalization step attempted to explain the different processing methods through text and then data visualization. Making sure that the user understands the impact this step has on their data is significant for downstream.

Select Normalization Method screen with a summary of AOIs and genes removed and retained, a dropdown showing the recommended method, and a chart comparing normalization methods.
Select Normalization Method screen with a summary of AOIs and genes removed and retained, a dropdown showing the recommended method, and a chart comparing normalization methods.

As the last step before analysis, it shows a summary of the experiment and the option to choose a normalization method, with the system suggesting one if the user is unsure.

As the last step before analysis, it shows a summary of the experiment and the option to choose a normalization method, with the system suggesting one if the user is unsure.

Outcomes

Outcomes

Before: researcher with a hand on her forehead, frustrated at a monitor of code and uneven box plots.

Before

After: the same researcher smiling at a monitor showing the new GeoMx QC screen with a histogram.

After

Before, spatial transcriptomics analysis had skewed results due to poor QC and unreliable normalization methods. This guided workflow would become a powerful tool for biologists and researchers.

Internally, the platform's internal design system was updated, providing a solid foundation for future design and development.

Before, spatial transcriptomics analysis had skewed results due to poor QC and unreliable normalization methods. This guided workflow would become a powerful tool for biologists and researchers.

Internally, the platform's internal design system was updated, providing a solid foundation for future design and development.

8

updated components

25+

New users

52 hours

Saved on design debt fixes

Reflections

Reflections

Simplified representation of the GeoMx QC screen, showing a clean layout of data and messaging.
What worked
What worked
  • Rich QC screens built on a proven guided workflow pattern and baseline components.


  • The product owner remarked on how clean the layout and data looked.

  • Rich QC screens built on a proven guided workflow pattern and baseline components.


  • The product owner remarked on how clean the layout and data looked.

Researcher at a desk next to their monitor and imaging device in the background, a pair of hands with a notebook and pen in the foreground.
I'd have done Differently
I'd have done Differently
  • Involve scientists and observe their workflows and working environments earlier.


  • Did most users work at stationary monitors or should we have accounted for mobile experiences too?

  • Involve scientists and observe their workflows and working environments earlier.


  • Did most users work at stationary monitors or should we have accounted for mobile experiences too?

Looping diagram: Prototype (wireframe) leads to Test (researcher at a laptop), then Refine (shapes and a cursor), then back to Prototype.
Moving Forward
Moving Forward
  • Pressure test working prototypes early and often with every sprint.


  • Interactive prototypes would be more cost-efficient to test than fully developed features.

  • Pressure test working prototypes early and often with every sprint.


  • Interactive prototypes would be more cost-efficient to test than fully developed features.

Designing workflows for complex domains relies on either deeply understanding the domain or collaborating with those who do.

Designing workflows for complex domains relies on either deeply understanding the domain or collaborating with those who do.

Next Project…

Next Project…

Helping fight prior Authorization Denials With AI

Leveraging AI design tools combined with usability testing to build a tool that helps patients understand and fight their health insurance denials.

Leveraging AI design tools combined with usability testing to build a tool that helps patients understand and fight their health insurance denials.

SecondLook Denial Letter Analyzer screen with a PDF upload area and a field to paste letter text.

Questing for a product designer?

Open to full-time roles and select projects

Questing for a product designer?

Open to full-time roles and select projects

Questing for a product designer?

Open to full-time roles and select projects