Concepts

REVISE reconstructs Spatially-inferred Virtual Cells (SVCs) from spatial transcriptomics data, tissue morphology, and matched single-cell RNA-seq references. The central idea is to choose the reconstruction route from the data’s resolution and dominant technical limitation, then let the unified pipeline resolve the correct strategy from revise/revise.yaml.

What Is an SVC?

An SVC is a reconstructed cell-level unit represented as an AnnData object. It carries inferred expression and/or spatial information that can be consumed by the case notebooks and downstream analysis services.

SVC kind	What it improves	When to use it
`sp-SVC`	Spatial localization and local tissue architecture.	Use this when high-definition ST already has rich spatial signal but suffers from segmentation or bin-to-cell assignment uncertainty.
`sc-SVC`	Molecular completeness and cell-state resolution.	Use this when an imaging or sequencing ST platform benefits from matched scRNA-seq references to restore missing genes or refine selected cell types.

Route Selection

REVISE routes by platform and confounding. In normal use, choose a profile first; override the platform/confounding pair only when validating a new route.

Profile	SVC	Platform and confounding	Typical question
`application_sp`	`sp-SVC`	`hST` + `bin2cell`	Can we refine Visium HD-style bins into biologically coherent cell-level spatial units?
`application_sc`	`sc-SVC`	`iST` + `segmentation`	Can we refine Xenium-style cell measurements with matched scRNA-seq and recover whole-transcriptome signals for a selected cell type?
`application_sc_sst`	`sc-SVC`	`sST` + `spot_size`	Can spot-level data be reconstructed at higher effective cellular resolution?
`benchmark_*`	`sp-SVC` or `sc-SVC`	`sim2real` + one of six confounding factors	Does reconstruction recover ground-truth SVCs under controlled segmentation, bin2cell, batch, spot-size, gene-panel, or dropout perturbations?

Confounding Factors

Current ST limitations addressed by REVISE — REVISE separates spatially heterogeneous limitations from spatially homogeneous limitations, then maps each family to a concrete benchmark or application route.

Family	Factors	REVISE route
Spatially heterogeneous	Segmentation artifacts and bin-to-cell assignment errors.	`sp-SVC` routes for hST and Sim2Real-ST segmentation/bin2cell benchmarks.
Spatially homogeneous	Spot size, batch effect, gene panel limitation, and gene dropout.	`sc-SVC` super-resolution and imputation routes.

Inputs and Outputs

Most runs need an ST file, a matched single-cell reference, and a writable output root. Benchmark runs also need the Sim2Real-ST ground-truth SVC file.

Input or output	Config key	Notes
Spatial transcriptomics data	`io.st_file`	`AnnData` file containing spatial measurements for the selected sample.
Single-cell reference	`io.sc_ref_file`	Matched scRNA-seq reference used by anchoring and local refinement.
Ground truth SVC	`io.gt_svc_file`	Required for benchmark evaluation.
Published application outputs	`sp_SVC.h5ad`, `sc_SVC_expr.h5ad`, `sc_SVC_spatial.h5ad`	Notebook-compatible copies produced by application wrappers.
Canonical run metadata	`merged_config.json`, `provenance.json`	Written for every unified run to make configuration and data fingerprints inspectable.

Where to Go Next

Run the shortest working path in Quick Start.
Learn the wrapper scripts and notebook outputs in Application Mode.
Inspect routing and override rules in Configuration.
Extend strategies or plugins using Architecture and API.