scviva-tools 0.1.5

Consolidated spatial transcriptomics analysis toolkit with variational inference methods

scviva-tools

Consolidated spatial transcriptomics analysis toolkit based on scvi-tools.

Provides seven spatial transcriptomics models and tools as first-class citizens:

• scVIVA — niche-aware representation learning (Levy et al., 2025)
• DestVI — multi-resolution cell-type deconvolution (Lopez et al., 2022)
• ResolVI — noise and bias correction for cellular-resolution ST (Ergen & Yosef, 2025)
• GIMVI — joint imputation of missing genes across paired scRNA-seq and spatial data (Lopez et al., 2019)
• Stereoscope — probabilistic cell-type deconvolution of spatial spots (Andersson et al., 2020)
• Tangram — mapping scRNA-seq data onto spatial coordinates (Biancalani et al., 2021)
• Harreman — metabolic exchange and cell-cell communication inference for spatial data (Etxezarreta-Arrastoa et al., 2025), integrating outputs from DestVI, ResolVI, and SCVIVA

See docs/user_guide/models for a full description of each model, and docs/tutorials for worked examples of every model above.

Installation

pip install scviva-tools

# Optional extras
pip install "scviva-tools[spatial]"    # SpatialData + squidpy integration
pip install "scviva-tools[rapids]"     # GPU-accelerated neighbor graphs (cuML/cuPy/cuGraph)
pip install "scviva-tools[tutorials]"  # jupyter, matplotlib, seaborn
pip install "scviva-tools[all]"        # everything above

See docs/installation.md for details, including a development install.

Quick Start

import scviva

# scVIVA
scviva.SCVIVA.setup_anndata(adata, layer="counts", spatial_key="spatial")
model = scviva.SCVIVA(adata)
model.train()

# DestVI
import scvi
scvi.model.CondSCVI.setup_anndata(sc_adata, labels_key="cell_type", layer="counts")
sc_model = scvi.model.CondSCVI(sc_adata)
sc_model.train()

scviva.DestVI.setup_anndata(st_adata, layer="counts")
st_model = scviva.DestVI.from_rna_model(st_adata, sc_model)
st_model.train()

# ResolVI
scviva.ResolVI.setup_anndata(adata, layer="counts", spatial_key="spatial")
model = scviva.ResolVI(adata)
model.train()

Models beyond scVIVA/DestVI/ResolVI (GIMVI, Stereoscope, Tangram) live under scviva.external, e.g. scviva.external.Tangram.

Downstream tools (scviva.tl / scviva.pl)

scviva.tl and scviva.pl mirror scanpy's sc.tl/sc.pl convention, resolving lazily to scviva.tools/scviva.plotting so heavy optional dependencies aren't imported eagerly:

# Harreman: metabolic exchange & cell-cell communication analysis, optionally
# integrating DestVI/ResolVI/SCVIVA outputs
scviva.tl.harreman.hs.compute_local_autocorrelation(adata, use_metabolic_genes=True)
scviva.pl.harreman.local_correlation_plot(adata)

scviva.tl.harreman itself exposes a scanpy-style sub-API: tl/hs/pp/ds/pl for tools/hotspot/preprocessing/datasets/plotting, plus a stateful HarremanAnalysis wrapper (scviva.tools.harreman.HarremanAnalysis) that can integrate DestVI/ResolVI/SCVIVA outputs.

Documentation

Full documentation, including the user guide, API reference, and tutorials, is under docs/. See CHANGELOG.md for a detailed history of recent changes.

References

See docs/references.md for full citations.

Copyright (c) 2026, Yosef Lab, Weizmann Institute of Science