scvi-tools

Name: scvi-tools
Author: LeonChaoX

698stars61forks

How to install scvi-tools?

leonchaox/qinyan-academic-skills/scvi-tools

$npx -y skills add LeonChaoX/qinyan-academic-skills --skill scvi-tools

Installs into the current project.

Summary

Deep generative models for single-cell omics. Use when you need probabilistic batch correction (scVI), transfer learning, differential expression with uncertainty, or multi-modal integration (TOTALVI, MultiVI). Best for advanced modeling, batch effects, multimodal data. For stand

Files · 1

View on GitHub

SKILL.md

1	# scvi-tools
2
3	## Overview
4
5	scvi-tools is a comprehensive Python framework for probabilistic models in single-cell genomics. Built on PyTorch and PyTorch Lightning, it provides deep generative models using variational inference for analyzing diverse single-cell data modalities.
6
7	## When to Use This Skill
8
9	Use this skill when:
10	- Analyzing single-cell RNA-seq data (dimensionality reduction, batch correction, integration)
11	- Working with single-cell ATAC-seq or chromatin accessibility data
12	- Integrating multimodal data (CITE-seq, multiome, paired/unpaired datasets)
13	- Analyzing spatial transcriptomics data (deconvolution, spatial mapping)
14	- Performing differential expression analysis on single-cell data
15	- Conducting cell type annotation or transfer learning tasks
16	- Working with specialized single-cell modalities (methylation, cytometry, RNA velocity)
17	- Building custom probabilistic models for single-cell analysis
18
19	## Core Capabilities
20
21	scvi-tools provides models organized by data modality:
22
23	### 1. Single-Cell RNA-seq Analysis
24	Core models for expression analysis, batch correction, and integration. See `references/models-scrna-seq.md` for:
25	- scVI: Unsupervised dimensionality reduction and batch correction
26	- scANVI: Semi-supervised cell type annotation and integration
27	- AUTOZI: Zero-inflation detection and modeling
28	- VeloVI: RNA velocity analysis
29	- contrastiveVI: Perturbation effect isolation
30
31	### 2. Chromatin Accessibility (ATAC-seq)
32	Models for analyzing single-cell chromatin data. See `references/models-atac-seq.md` for:
33	- PeakVI: Peak-based ATAC-seq analysis and integration
34	- PoissonVI: Quantitative fragment count modeling
35	- scBasset: Deep learning approach with motif analysis
36
37	### 3. Multimodal & Multi-omics Integration
38	Joint analysis of multiple data types. See `references/models-multimodal.md` for:
39	- totalVI: CITE-seq protein and RNA joint modeling
40	- MultiVI: Paired and unpaired multi-omic integration
41	- MrVI: Multi-resolution cross-sample analysis
42
43	### 4. Spatial Transcriptomics
44	Spatially-resolved transcriptomics analysis. See `references/models-spatial.md` for:
45	- DestVI: Multi-resolution spatial deconvolution
46	- Stereoscope: Cell type deconvolution
47	- Tangram: Spatial mapping and integration
48	- scVIVA: Cell-environment relationship analysis
49
50	### 5. Specialized Modalities
51	Additional specialized analysis tools. See `references/models-specialized.md` for:
52	- MethylVI/MethylANVI: Single-cell methylation analysis
53	- CytoVI: Flow/mass cytometry batch correction
54	- Solo: Doublet detection
55	- CellAssign: Marker-based cell type annotation
56
57	## Typical Workflow
58
59	All scvi-tools models follow a consistent API pattern:
60
61	```python
62	# 1. Load and preprocess data (AnnData format)
63	import scvi
64	import scanpy as sc
65
66	adata = scvi.data.heart_cell_atlas_subsampled()
67	sc.pp.filter_genes(adata, min_counts=3)
68	sc.pp.highly_variable_genes(adata, n_top_genes=1200)
69
70	# 2. Register data with model (specify layers, covariates)
71	scvi.model.SCVI.setup_anndata(
72	adata,
73	layer="counts", # Use raw counts, not log-normalized
74	batch_key="batch",
75	categorical_covariate_keys=["donor"],
76	continuous_covariate_keys=["percent_mito"]
77	)
78
79	# 3. Create and train model
80	model = scvi.model.SCVI(adata)
81	model.train()
82
83	# 4. Extract latent representations and normalized values
84	latent = model.get_latent_representation()
85	normalized = model.get_normalized_expression(library_size=1e4)
86
87	# 5. Store in AnnData for downstream analysis
88	adata.obsm["X_scVI"] = latent
89	adata.layers["scvi_normalized"] = normalized
90
91	# 6. Downstream analysis with scanpy
92	sc.pp.neighbors(adata, use_rep="X_scVI")
93	sc.tl.umap(adata)
94	sc.tl.leiden(adata)
95	```
96
97	Key Design Principles:
98	- Raw counts required: Models expect unnormalized count data for optimal performance
99	- Unified API: Consistent interface across all models (setup → train → extract)
100	- AnnData-centric: Seamless integration with the scanpy ecosystem
101	- GPU acceleration: Automatic utilization of available GPUs
102	- Batch correction: Handle technical variation through covariate registration
103
104	## Common Analysis Tasks
105
106	### Differential Expression
107	Probabilistic DE analysis using the learned generative models:
108
109	```python
110	de_results = model.differential_expression(
111	groupby="cell_type",
112	group1="TypeA",
113	group2="TypeB",
114	mode="change", # Use composite hypothesis testing
115	delta=0.25 # Minimum effect size threshold
116	)
117	```
118
119	See `references/differential-expression.md` for detailed methodology and interpretation.
120
121	### Mode

People who liked this also viewed

6 picks

Type

scvi-tools

698stars61forks

View on GitHub

How to install scvi-tools?

leonchaox/qinyan-academic-skills/scvi-tools

$npx -y skills add LeonChaoX/qinyan-academic-skills --skill scvi-tools

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# scvi-tools
2
3	## Overview
4
5	scvi-tools is a comprehensive Python framework for probabilistic models in single-cell genomics. Built on PyTorch and PyTorch Lightning, it provides deep generative models using variational inference for analyzing diverse single-cell data modalities.
6
7	## When to Use This Skill
8
9	Use this skill when:
10	- Analyzing single-cell RNA-seq data (dimensionality reduction, batch correction, integration)
11	- Working with single-cell ATAC-seq or chromatin accessibility data
12	- Integrating multimodal data (CITE-seq, multiome, paired/unpaired datasets)
13	- Analyzing spatial transcriptomics data (deconvolution, spatial mapping)
14	- Performing differential expression analysis on single-cell data
15	- Conducting cell type annotation or transfer learning tasks
16	- Working with specialized single-cell modalities (methylation, cytometry, RNA velocity)
17	- Building custom probabilistic models for single-cell analysis
18
19	## Core Capabilities
20
21	scvi-tools provides models organized by data modality:
22
23	### 1. Single-Cell RNA-seq Analysis
24	Core models for expression analysis, batch correction, and integration. See `references/models-scrna-seq.md` for:
25	- scVI: Unsupervised dimensionality reduction and batch correction
26	- scANVI: Semi-supervised cell type annotation and integration
27	- AUTOZI: Zero-inflation detection and modeling
28	- VeloVI: RNA velocity analysis
29	- contrastiveVI: Perturbation effect isolation
30
31	### 2. Chromatin Accessibility (ATAC-seq)
32	Models for analyzing single-cell chromatin data. See `references/models-atac-seq.md` for:
33	- PeakVI: Peak-based ATAC-seq analysis and integration
34	- PoissonVI: Quantitative fragment count modeling
35	- scBasset: Deep learning approach with motif analysis
36
37	### 3. Multimodal & Multi-omics Integration
38	Joint analysis of multiple data types. See `references/models-multimodal.md` for:
39	- totalVI: CITE-seq protein and RNA joint modeling
40	- MultiVI: Paired and unpaired multi-omic integration
41	- MrVI: Multi-resolution cross-sample analysis
42
43	### 4. Spatial Transcriptomics
44	Spatially-resolved transcriptomics analysis. See `references/models-spatial.md` for:
45	- DestVI: Multi-resolution spatial deconvolution
46	- Stereoscope: Cell type deconvolution
47	- Tangram: Spatial mapping and integration
48	- scVIVA: Cell-environment relationship analysis
49
50	### 5. Specialized Modalities
51	Additional specialized analysis tools. See `references/models-specialized.md` for:
52	- MethylVI/MethylANVI: Single-cell methylation analysis
53	- CytoVI: Flow/mass cytometry batch correction
54	- Solo: Doublet detection
55	- CellAssign: Marker-based cell type annotation
56
57	## Typical Workflow
58
59	All scvi-tools models follow a consistent API pattern:
60
61	```python
62	# 1. Load and preprocess data (AnnData format)
63	import scvi
64	import scanpy as sc
65
66	adata = scvi.data.heart_cell_atlas_subsampled()
67	sc.pp.filter_genes(adata, min_counts=3)
68	sc.pp.highly_variable_genes(adata, n_top_genes=1200)
69
70	# 2. Register data with model (specify layers, covariates)
71	scvi.model.SCVI.setup_anndata(
72	adata,
73	layer="counts", # Use raw counts, not log-normalized
74	batch_key="batch",
75	categorical_covariate_keys=["donor"],
76	continuous_covariate_keys=["percent_mito"]
77	)
78
79	# 3. Create and train model
80	model = scvi.model.SCVI(adata)
81	model.train()
82
83	# 4. Extract latent representations and normalized values
84	latent = model.get_latent_representation()
85	normalized = model.get_normalized_expression(library_size=1e4)
86
87	# 5. Store in AnnData for downstream analysis
88	adata.obsm["X_scVI"] = latent
89	adata.layers["scvi_normalized"] = normalized
90
91	# 6. Downstream analysis with scanpy
92	sc.pp.neighbors(adata, use_rep="X_scVI")
93	sc.tl.umap(adata)
94	sc.tl.leiden(adata)
95	```
96
97	Key Design Principles:
98	- Raw counts required: Models expect unnormalized count data for optimal performance
99	- Unified API: Consistent interface across all models (setup → train → extract)
100	- AnnData-centric: Seamless integration with the scanpy ecosystem
101	- GPU acceleration: Automatic utilization of available GPUs
102	- Batch correction: Handle technical variation through covariate registration
103
104	## Common Analysis Tasks
105
106	### Differential Expression
107	Probabilistic DE analysis using the learned generative models:
108
109	```python
110	de_results = model.differential_expression(
111	groupby="cell_type",
112	group1="TypeA",
113	group2="TypeB",
114	mode="change", # Use composite hypothesis testing
115	delta=0.25 # Minimum effect size threshold
116	)
117	```
118
119	See `references/differential-expression.md` for detailed methodology and interpretation.
120
121	### Mode

People who liked this also viewed

6 picks

Type