REVISE Documentation

REconstruction via Vision-integrated Spatial Estimation

Spatially inferred virtual cells from ST, histology, and scRNA-seq.

REVISE reconstructs Spatially-inferred Virtual Cells (SVCs) by routing each dataset through a config-driven topology-aware optimal transport pipeline. The documentation is organized around the decisions users make first: which SVC mode to run, which platform/confounding route applies, and how outputs flow into benchmark or application notebooks.

Start with a route Understand SVCs See the architecture

Resources

Overview of the REVISE framework

REVISE combines spatial transcriptomics, histology, and single-cell reference information to reconstruct SVCs for benchmark and real-data application workflows.

Choose Your Path

Concepts

Know what REVISE reconstructs

Compare sp-SVC and sc-SVC, map platforms to confounding factors, and learn the expected inputs and outputs before running a case.

 Quick start

Run the shortest useful example

Install the package, pick a route, run a wrapper script, or call the unified Python API directly.

 Applications

Reconstruct real tissue sections

Use sp-SVC for Visium HD-style spatial refinement or sc-SVC for Xenium/Visium molecular completion and downstream biology.

 Benchmarks

Reproduce Sim2Real-ST evaluations

Run the six confounding-factor families and inspect normalized PCC, SSIM, MSE, and NRMSE outputs.

 Gallery

Browse rendered case notebooks

Open paper-facing application notebooks for sp-SVC and sc-SVC cases, including Xenium cell-type analyses and Visium HD reconstruction.

 Configuration

Control profiles and routes

Understand how defaults, profiles, runtime overrides, IO overrides, and dotted set-overrides merge into a single run configuration.

 API

Extend the unified engine

Use REVISEPipeline, SVC result carriers, strategy registries, plugin registries, and analysis services.

Two SVC Modes

Mode

Best for

Typical platforms

Primary outputs

sp-SVC

Spatial refinement when high-definition ST is affected by segmentation or bin-to-cell assignment artifacts.

hST such as Visium HD; Sim2Real-ST segmentation and bin2cell benchmarks.

sp_SVC.h5ad and canonical artifacts/sp_svc.h5ad.

sc-SVC

Molecular completion and cell-state refinement when ST measures limited genes or spot-level mixtures.

iST/sST such as Xenium and Visium; super-resolution and imputation benchmarks.

sc_SVC_expr.h5ad, sc_SVC_spatial.h5ad, and canonical artifacts.

Core Execution Model

All modern execution flows through one public engine:

  1. revise.framework.REVISEPipeline.run()

  2. revise/revise.yaml profiles and runtime/io overrides

  3. revise.recon.pipeline.UnifiedReconstructionPipeline

  4. backend strategy and plugin registries in revise/backend/

The fixed lifecycle is documented in Architecture; practical profile and override examples are documented in Configuration.