EP16: EXPLORING SPATIALX: A STEP-BY-STEP GUIDE TO ANALYZING MULTIMODAL SPATIAL DATA

Main ideas / lessons conveyed

• Purpose of the session: Provide a step-by-step overview of how to analyze multimodal spatial data using SpaceX, specifically targeting MONTI model spatial data (as stated).
• Why SpaceX was built: Existing “spal/spatial” workflows were considered insufficient for key spatial-data challenges.
• Three core challenges SpaceX aims to solve:
  1. Integrate multiple spatial technologies (side-by-side comparison and combined workflows).
  2. Improve cell type annotation quality (automate while supporting validation).
  3. Improve downstream analysis using integrated/matched properties of different technologies.

Methodology / workflow presented

0) Pre-session / onboarding

After the session, viewers can:

• Find more information and book a meeting via a website link (shown as by.com SpaceX in the subtitles).
• Use a public dataset option on SpaceX (described as “space r” / “space x r” in the subtitles) to explore publicly available tissues/diseases if they lack in-house data.

1) Challenge 1: Integrate multiple spatial technologies

Problem context (as described)

• In spatial research, different papers often use multiple technologies on the same kind of sample to find the best fit.
• Technologies can differ in:
  • Resolution (e.g., single-cell-like vs lower resolution / pixelated).
  • Gene panel size (e.g., smaller vs much larger number of genes).
• As a result, gene expression/localization may appear different across technologies.

Core solution approach in SpaceX / Spacewise

• Support many technologies on one platform (examples mentioned include 10x Genomics / Visium, NanoString, Vantage, Lunaphore, and others).
• Let users:
  • View one technology at a time in older workflows,
  • While SpaceX adds the ability to analyze side-by-side across technologies.

Demonstrated capability: side-by-side gene expression comparison

• Example dataset: colon cancer
• Marker used in demo: PIGR
• Comparison: Zennium vs Visium HD
  • Zennium: higher resolution; can localize transcripts at (near) single-cell resolution; smaller gene panel.
  • Visium HD: lower spatial resolution (more pixelated); detects tens of thousands of genes for stronger downstream analysis.
• Marker localization was visualized to show how PIGR maps to ring-like structures (described as goblet-cell features).

Useful UI feature: Unified Zoom mode

• A layout switcher allows viewing:
  • 1 image, 2 images, or up to 4 images simultaneously.
• Provides consistent zooming across images without repeating steps.

---

2) Challenge 2: Cell type annotation (automate + validate)

Problem context (as described)

• Cell type annotation is commonly:
  • Manual (specialists), or
  • Automatic (models trained on large databases).
• SpaceX aims to combine these modes: automate predictions while keeping a validation/interpretation layer.

Solution component: Meta Reference annotation tool

• Presented as a human-in-the-loop style tool (as described).
• Uses a large reference database (figures given in subtitles):
  • ~2,000 datasets
  • ~40,000+ cells
  • ~1,000 cell types
• Key behavior:
  • Does not output only one final label.
  • Returns multiple annotation candidates per cluster, because a cluster can include multiple cell types.

What the results page shows (as demonstrated)

For each cluster:

• Major cell types predicted
• Sub-cell types
• Agreement weight / agreement rate
  • Defined as the proportion of the reference library that agrees with the prediction
• Core markers used for prediction
• Marker genes to assess plausibility

An info/details tab includes links back to:

• Standardized cell types
• Author’s annotation
• Specific studies providing marker evidence

Interpretation guidance provided in the demo

• If agreement is low or clusters appear mixed:
  • consider adjusting clustering resolution to obtain purer cell-type clusters.
• Validation emphasized:
  • Users can compare standardized terms vs author-specific annotations to preserve interpretive nuance (e.g., “goblet” vs “immature goblet”).

---

3) Challenge 2b (supported concept): Use integrated technologies to confirm spatial patterns

• Cross-check localization patterns using multiple technologies:
  • Zennium can show finer ring structures more clearly.
  • Visium may not show ring shape sharply, but can still confirm that the same cell types localize to the same region.
• A follow-up zoom example highlighted intracellular vs ring-region transcript capture intensity (darker red = more transcripts; lighter yellow = lower).

---

4) Challenge 3: Improve downstream analysis using spatial context and gene panel differences

Problem context (as described)

• Different technologies yield different data qualities (resolution and gene panel size).
• Therefore, technologies may be better suited for different downstream goals.

Downstream tool: Differential Gene Expression (DGE)

• Demo: Compare goblet cells vs fibroblasts across Zennium and Visium HD.
• Key shown:
  • Both technologies can find the same up/down genes (examples included PIGR and TAGLN).
  • Visium identifies more upregulated/downregulated genes due to higher detection capacity.
• Additional option:
  • Compare cell types across technologies (e.g., goblet in one tech vs goblet in another) as cross-tech comparisons.

New/emphasized tool: Region segmentation

• Concept:
  • Segmentation/clustering uses both:
    • gene expression, and
    • spatial context (where expression occurs, and whether similar-expression regions are near or far).
• Why it matters:
  • Claimed to improve spatial biology questions by producing better segmentation.

Region segmentation workflow + downstream pathway enrichment

• Steps (as demonstrated conceptually):
  1. Segment the tissue into region(s).
  2. Select a segment (e.g., “goblet”).
  3. Retrieve the segment-specific gene panel.
  4. Run pathway enrichment on those genes.
• Technology-dependent effect:
  • Zennium segment:
    • smaller gene panel (~“around 40 genes” stated in subtitles),
    • resulting in weaker/less informative enrichment.
  • Visium HD segment:
    • larger gene set (~“around 100+” implied/garbled; Visium overall described earlier as around ten thousand genes),
    • producing richer, more related pathway signals.
• Example pathway findings (colon goblet/cancer context):
  • enrichment of goblet cell carcinoma
  • enrichment of small intestine cancer-related pathways
• Conclusion:
  • Comparing both technologies helps produce better answers because gene panel size and spatial resolution tradeoffs differ.

Platform capabilities / future directions mentioned

• SpaceX beta is under active development, with upcoming features:
  • More visualization tools
  • New cell type annotation method (Meta Reference already demonstrated)
  • New analysis tools (including region segmentation, discussed)
• Data migration:
  • If data exists in “bio lens” (subtitles), users can migrate to SpaceX with one button (“migrate from lens”).
• Roadmap/poll input:
  • Suggested future advanced analyses:
    • Neighborhood analysis
    • Cell (or sell) interaction analysis
• Upcoming session:
  • Next week will dive deeper into a biological story using SpaceX.

Speakers / sources featured

• Unspecified speaker(s): multiple presenters referenced as “I” and “we” (names not fully captured in subtitles).
• Julie: main demonstrator/presenter of live platform features, including:
  • Meta Reference tour
  • Results explanation
  • Gene expression examples
  • Region segmentation explanations
• Sarah ATO / Sarah: referenced as an author/source for a paper example using multiple spatial technologies (exact full citation not provided in subtitles).