Bio+MedVis Challenge @ IEEE VIS 2024

Redesign Challenge: Redesign an Existing Visualization

Biological Background and Data Description

In biology, structure and function are tightly linked, and the biology of cells in tissues is no exception. A key challenge for cell biologists lies in understanding the spatial patterns characteristic of health and disease. While historical single-cell profiling techniques have required tissue dissociation, a cutting-edge area of cell biology research is the development and application of technologies that measure molecular states within cells while preserving the spatial context that is critical to establishing a localized picture of tissue health.

A key cellular measurement technique is called transcriptomics, which provides insight into cell types (e.g., immune cell vs. muscle cell), cell states (e.g., response to stimuli), and information about proteins that might be produced downstream (i.e., via translation). Transcriptomics techniques acquire this information by measuring gene expression: the abundance of RNA transcripts corresponding to each gene in each cell. As the name implies, spatial transcriptomics methods capture gene expression and spatial coordinates simultaneously. Resolutions of spatial transcriptomics methods can vary from subcellular, to single-cell, to “spots” that represent multiple cells. There is typically a tradeoff between the spatial resolution and the number of genes that can be measured by a given technology.

VISIUM is a spatial transcriptomics technology developed by 10x Genomics. It allows researchers to measure gene expression and spatial organization of tissues simultaneously. For a given tissue slice, VISIUM is able to measure transcripts genome-wide (e.g., ~20,000 genes for human samples) at a resolution of approximately 5-10 cells per spot. This enables researchers to understand how gene expression patterns vary across different regions of a tissue sample. This technology has applications in various fields such as cancer research, developmental biology, neuroscience, and more, allowing researchers to gain deeper insights into the molecular mechanisms underlying complex biological processes within tissues.

While each VISIUM spot represents the gene expression of multiple cells, researchers are typically interested in interpreting the data at the cell type level (e.g., linking expression differences to particular cell types that are characterized in the biological literature). As a result, computational methods have been developed to deconvolute each VISIUM spot into a distribution of (predicted) cell-type proportions. Biologists can then use the deconvolution results to link their findings to their existing cell type knowledge.

These deconvolutional methods thus produce data that, on a per-spot basis, represents proportional cell-type membership (e.g. 30% Cell Type 1, 50% Cell Type 2, 20% Cell Type 3). Visualizing these data in the context of the H&E stained tissue image provides crucial spatial context, enabling domain experts to explore deconvolution results within the intricate cellular architecture and tissue morphology. One common technique for visualizing these data is to superimpose a pie chart on each spot location representing the cell types represented. Pie charts and their limitations have been well explored by the visualization community. Moreover, superimposing them on the image occludes the underlying tissue and prevents important contextual inspection of a given spot.

Redesign Challenge Task

For this redesign task, we challenge participants to propose an alternative visualization approach that links the spot cell-type proportion to the H&E image. Submissions should consider visualization theory and principles.

Sketching and prototyping as well as fully interactive solutions will be accepted. Creativity and novelty will be considered in the evaluation.

Data and Documentation

Example Dataset

https://drive.google.com/drive/folders/1t6aeMDh2l067_6DEMFyovtRO5swZieBF?usp=sharing

Data Description

• Genes.csv - List of genes in the feature matrix
• FeatureMatrix.mtx - Feature matrix, stored as a sparse matrix
• ClusterGeneExpression.csv - Gene expression per cell-type cluster
• SpotClusterMembership.csv - Cell type proportions per spot
• SpotPositions.csv - Spot positions and radiuses
• Images/scalefactors_json.json - Scale factors between spot positions
• Images/tissue_hires_image.png - H&E image

Additional Data and Resources

• Visium data from Kleshchevnikov et al., 2022: https://www.ebi.ac.uk/biostudies/arrayexpress/studies/E-MTAB-11114, filter by sample ST8059049
• Summary: https://ftp.ebi.ac.uk/biostudies/fire/E-MTAB-/114/E-MTAB-11114/Files/ST8059049_web_summary.html
• Vitessce demo of this data: http://vitessce.io/#?dataset=spatialdata-visium
• Vitessce documentation & information: http://vitessce.io/docs/

References

Kleshchevnikov, V., Shmatko, A., Dann, E. et al. Cell2location maps fine-grained cell types in spatial transcriptomics. Nat Biotechnol 40, 661–671 (2022). https://doi.org/10.1038/s41587-021-01139-4

Miller, B.F., Huang, F., Atta, L. et al. Reference-free cell type deconvolution of multi-cellular pixel-resolution spatially resolved transcriptomics data. Nat Commun 13, 2339 (2022). https://doi.org/10.1038/s41467-022-30033-z

3D Microscopy Imaging Challenge: From a RAW imaging volume to biological findings

Biological Background and Data Description

Highly multiplexed tissue imaging methods, such as Cyclic Immunofluorescence (CycIF), which allow for the analysis of over 30 biomarkers on a single tissue section, are essential tools for scientists to investigate the subcellular complexities of cancer [1]. Indeed, CyCIF has been instrumental in revealing immune-tumor interactions and the progression of melanoma at single-cell precision [2]. More recently, researchers have extended these techniques to image volumes, allowing for an even more comprehensive analysis of the diverse cell types and states within the tumor microenvironment and their spatial interactions [3]. However, effective visualization and investigation of these volumes is difficult, given their size and dimensionality. Scientists now seek new methods that can address the challenges of data occlusion and improve the process of exploring and discovering these hard-to-see interactions. The first step towards this—and the focus of this BioMedVis Challenge—is to develop algorithms that optimize camera views of Regions of Interest (ROIs) that feature important subsets of biomarkers.

Challenge Task

The starting Jupyter Notebook presents one ROI with 4 channels of interest in the dataset and a second ROI with 6 channels of interest. It is your task to find the best possible camera view configurations for a given ROI that 1) shows maximum spatial interaction of selected channels and 2) minimizes data occlusion. You can start with the first two areas given as X, Y, Z coordinates.

The following settings must be defined for each view configuration:

• Camera:
  • Zoom Level
• Volume Settings:
  • Translation (X, Y, Z)
  • Rotation (X, Y, Z)
• Channels:
  • Selection, Color, and Value Range

As a more advanced alternative, you can also try to adapt the shader code given in Vitessce to allow for more elaborate rendering operations (e.g., adaptive transparency to highlight structures of interest and reduce occlusion along the view ray).

Optimization Criteria:

You should optimize your algorithm for finding potential view configurations by having the following criteria in mind:

• Find the best possible view for the target structure of interest with minimal occlusion
• You should activate at least two channels (but you can activate as many as you see fit)
• For one target you can find multiple views
  • Find the best sequence of views to show the regions of interest (ROIs) in the best possible way

Additionally, you can also:

• Find a way to navigate between the view configurations using animation
• Find a creative solution to make the ROIs visible while maintaining the context of the volume

Data and Documentation

The data volume included in this challenge represents a 194x5508x10908 volume of cancerous tissue belonging to a patient suffering from metastatic melanoma. Scientists have identified “immune niches” in this tissue, which contain specific interactions between immune cells of different types and states. Included in the data:

• 70 channels
• 2 ROI’s as examples in Jupyter

Also provided is a Jupyter Notebook (link/details), which includes all necessary starting code and information:

BioMedVis Challenge 2024.ipynb

References

[1] Lin JR, Fallahi-Sichani M, Chen JY, Sorger PK. Cyclic Immunofluorescence (CycIF), A Highly Multiplexed Method for Single-cell Imaging. Curr Protoc Chem Biol. 2016 Dec 7;8(4):251-264. doi: 10.1002/cpch.14. PMID: 27925668; PMCID: PMC5233430.

[2] Ajit J. Nirmal, Zoltan Maliga, Tuulia Vallius, Brian Quattrochi, Alyce A. Chen, Connor A. Jacobson, Roxanne J. Pelletier, Clarence Yapp, Raquel Arias-Camison, Yu-An Chen, Christine G. Lian, George F. Murphy, Sandro Santagata, Peter K. Sorger; The Spatial Landscape of Progression and Immunoediting in Primary Melanoma at Single-Cell Resolution. Cancer Discov 1 June 2022; 12 (6): 1518–1541. doi: 10.1158/2159-8290.CD-21-1357

[3] Yapp C, Nirmal AJ, Zhou F, Maliga Z, Tefft JB, Llopis PM, Murphy GF, Lian CG, Danuser G, Santagata S, Sorger PK; Human Tumour Atlas Network. Multiplexed 3D Analysis of Immune States and Niches in Human Tissue. bioRxiv [Preprint]. 2024 Mar 28:2023.11.10.566670. doi: 10.1101/2023.11.10.566670. PMID: 38014052; PMCID: PMC10680601.

Submission

Submissions will be considered for talk or poster presentations. Please send a two-page PDF abstract with up to 5 additional figures and a draft of your proposed poster (max 10MB) to PCS: new.precisionconference.com/submissions. The abstract should include:

• aspects of the figure identified as needing improvement or clarification,
• justification of encoding and design choices,
• at least one or more images of your design
• optional: a video or screencast to explain the visual encoding

Selected submissions will be invited for talk presentations during the Bio+MedVis session at the IEEE VIS 2024 conference.

Evaluation of Submissions

All submissions will be thoroughly reviewed by at least two reviewers, coming from the challenge chairs and selected domain experts. All accepted submissions will be published in the conference proceedings. Winning designs may be invited to become a plugin or extension to Vitessce.

Program

See the program for more details.

Important dates

• Submission: August 23, 2024
• Notification: August 29, 2024
• Camera-ready version: September 5, 2024
• Bio+MedVis Challenge event: October 13, 2024

Questions?

Please feel free to ask questions at: biovis_challenge@ieeevis.org.

The chairs of the Bio+MedVis Challenge @ IEEE VIS 2024:

• Barbora Kozlikova, Masaryk University, Czech Republic
• Nils Gehlenborg, Harvard Medical School, USA
• Laura Garrison, University of Bergen, Norway
• Eric Mörth, Harvard Medical School, USA
• Simon Warchol, Harvard University, USA
• Morgan Turner, Harvard Medical School, USA