Skip to content

sosta

Spatial Omics STructure Analysis — a Bioconductor package that reconstructs anatomically relevant tissue structures from spatial omics data (using molecular features or cell-type labels), then computes structure-level quantitative metrics — bridging the gap between cell-level spatial-omics analysis and tissue-architecture readouts.

Talk at GBCC2025

  • Speakers: Samuel Gunz, Mark D Robinson (University of Zurich)
  • Day / session: Day 3 (Wed Jun 25, 2025) — Oral Session 3, Grace Auditorium (chair: Lambda Moses)
  • Talk title: "sosta: a framework to analyze anatomical structures from spatial omics data"
  • Slides (Zenodo DOI): TBD — Zenodo deposits typically 2–4 weeks post-conference
  • Video: GBCC2025 playlist
  • Abstract / schedule: GBCC2025 program

What it does

Most spatial-omics tooling stops at the cell — segmentation, neighborhood graphs, niche detection. sosta goes one level up: take a spatial-omics object and reconstruct anatomical structures (e.g. islets, glomeruli, tumor regions, lobules) from molecular features or cell-type labels, then compute morphology and composition metrics on the resulting structures. That gives downstream analysis a tissue-architecture handle that doesn't require external segmentation masks.

How it works

Core idea. Density-based reconstruction of multicellular structures: starting from labeled cell coordinates (a cell type, or a molecular-feature threshold), sosta builds a spatial density field across the tissue and converts the density landscape into polygonal structure boundaries, which then become first-class objects for downstream metric calculation.

Inputs / outputs. Input: a SpatialExperiment with single-cell coordinates, an image identifier, and a cell-type (or molecular-feature) label whose spatial accumulation defines the target structure. Output: an sf (simple-features) polygon set representing reconstructed structures plus two metric tables — structure-level (shape descriptors, area, cell-type composition) and cell-level (distance-to-border, inside/outside boundary classification).

Key innovation. The Gunz/Crowell/Robinson preprint (bioRxiv 2025.10.13.682065) frames the gap explicitly: existing spatial-omics methods analyze "spatial arrangements of single cells, even though biological function often emerges from multicellular arrangements." sosta is the structure-level companion — it reconstructs anatomical regions directly from molecular/cell-type signal rather than requiring an external segmentation mask (CellPose, Mesmer, manual annotation). Direct head-to-head benchmark tools not enumerated in vignette excerpt.

Parameters worth knowing. The cell-type or feature label that defines the structure (e.g. islet cells for pancreatic islets). Bandwidth/smoothing for the density field. Density threshold for boundary extraction. Minimum structure size to filter noise. Exact parameter names and defaults not surfaced in vignette excerpt; needs talk slides.

Canonical example. Damond et al. (2019) IMC pancreatic-islets dataset — type-1-diabetes donors versus healthy controls across multiple images per donor. Pipeline: reconstructStructures builds islet polygons from islet-cell coordinates; getStructureMetrics computes per-islet shape and composition; mixed-effects models (lme4/lmerTest) compare metrics across disease groups while accounting for donor-level correlation. The overview vignette uses a simulated sostaSPE (5,641 cells, 3 images, 3 cell types A/B/C) as the introductory worked example.

Where it fits in the corpus

  • AACR 2026: axis single-cell-spatial-omics — structural-level readouts complement the cell-level dossiers (Cell2Location, etc.)
  • spatialFDA (entry) — same UZH / Robinson lab; spatialFDA tests differences in spatial summary functions, sosta produces the structures those functions can be evaluated on
  • SpatialData (entry) — sosta operates on the SpatialExperiment / SpatialData container family
  • OSTA (entry) — community textbook covering the broader spatial-omics methods stack
  • ISMB 2026: scaffold

Notes

Robinson lab (UZH) is consolidating a coherent spatial-omics methods stack — DESpace and spatialFDA at EuroBioC2025, now sosta at GBCC2025. The pancreatic-islets IMC vignette is a good showcase: islet reconstruction is a textbook anatomical-structure problem, and getting it for free out of the cell-level data is the value proposition.