scvi-tools/tabula-sapiens-fat-scanvi

ScANVI is a variational inference model for single-cell RNA-seq data that can learn an underlying latent space, integrate technical batches and impute dropouts. In addition, to scVI, ScANVI is a semi-supervised model that can leverage labeled data to learn a cell-type classifier in the latent space and afterward predict cell types of new data. The learned low-dimensional latent representation of the data can be used for visualization and clustering.

scANVI takes as input a scRNA-seq gene expression matrix with cells and genes as well as a cell-type annotation for a subset of cells. We provide an extensive user guide.

• See our original manuscript for further details of the model: scANVI manuscript.
• See our manuscript on scvi-hub how to leverage pre-trained models.

This model can be used for fine tuning on new data using our Arches framework: Arches tutorial.

Model Description

Tabula Sapiens is a benchmark, first-draft human cell atlas of nearly 500,000 cells from 24 organs of 15 normal human subjects.

Metrics

We provide here key performance metrics for the uploaded model, if provided by the data uploader.

Coefficient of variation

The cell-wise coefficient of variation summarizes how well variation between different cells is preserved by the generated model expression. Below a squared Pearson correlation coefficient of 0.4 , we would recommend not to use generated data for downstream analysis, while the generated latent space might still be useful for analysis.

Cell-wise Coefficient of Variation:

Metric	Training Value	Validation Value
Mean Absolute Error	1.32	1.33
Pearson Correlation	0.83	0.83
Spearman Correlation	0.81	0.82
R² (R-Squared)	0.61	0.59

The gene-wise coefficient of variation summarizes how well variation between different genes is preserved by the generated model expression. This value is usually quite high.

Gene-wise Coefficient of Variation:

Metric	Training Value
Mean Absolute Error	18.21
Pearson Correlation	0.65
Spearman Correlation	0.69
R² (R-Squared)	-1.21

Differential expression metric

The differential expression metric provides a summary of the differential expression analysis between cell types or input clusters. We provide here the F1-score, Pearson Correlation Coefficient of Log-Foldchanges, Spearman Correlation Coefficient, and Area Under the Precision Recall Curve (AUPRC) for the differential expression analysis using Wilcoxon Rank Sum test for each cell-type.

Differential expression:

Index	gene_f1	lfc_mae	lfc_pearson	lfc_spearman	roc_auc	pr_auc	n_cells
fibroblast	0.96	0.40	0.84	0.98	0.35	0.92	8459.00
endothelial cell	0.89	0.59	0.80	0.97	0.43	0.92	3748.00
macrophage	0.86	0.68	0.74	0.96	0.41	0.92	2807.00
T cell	0.94	1.51	0.65	0.89	0.26	0.86	1715.00
mesenchymal stem cell	0.91	1.03	0.71	0.94	0.41	0.89	1311.00
myofibroblast cell	0.95	1.58	0.69	0.90	0.28	0.82	982.00
mature NK T cell	0.89	2.42	0.63	0.85	0.29	0.81	572.00
neutrophil	0.89	3.08	0.69	0.78	0.20	0.92	440.00
leukocyte	0.52	5.05	0.51	0.68	0.17	0.73	160.00
plasma cell	0.69	4.79	0.56	0.60	0.25	0.75	26.00
B cell	0.25	10.46	0.34	0.37	0.20	0.72	23.00
smooth muscle cell	0.77	5.36	0.58	0.55	0.32	0.65	16.00

Model Properties

We provide here key parameters used to setup and train the model.

Model Parameters

These provide the settings to setup the original model:

{
    "n_hidden": 128,
    "n_latent": 20,
    "n_layers": 3,
    "dropout_rate": 0.05,
    "dispersion": "gene",
    "gene_likelihood": "nb",
    "linear_classifier": false,
    "latent_distribution": "normal",
    "use_batch_norm": "none",
    "use_layer_norm": "both",
    "encode_covariates": true
}

Setup Data Arguments

Arguments passed to setup_anndata of the original model:

{
    "labels_key": "cell_ontology_class",
    "unlabeled_category": "unknown",
    "layer": null,
    "batch_key": "donor_assay",
    "size_factor_key": null,
    "categorical_covariate_keys": null,
    "continuous_covariate_keys": null,
    "use_minified": false
}

Data Registry

Registry elements for AnnData manager:

Registry Key	scvi-tools Location
X	adata.X
batch	adata.obs['_scvi_batch']
labels	adata.obs['_scvi_labels']
latent_qzm	adata.obsm['scanvi_latent_qzm']
latent_qzv	adata.obsm['scanvi_latent_qzv']
minify_type	adata.uns['_scvi_adata_minify_type']
observed_lib_size	adata.obs['observed_lib_size']

• Data is Minified: False

Summary Statistics

Summary Stat Key	Value
n_batch	3
n_cells	20259
n_extra_categorical_covs	0
n_extra_continuous_covs	0
n_labels	13
n_latent_qzm	20
n_latent_qzv	20
n_vars	3000

Training

Training data url: Not provided by uploader

If provided by the original uploader, for those interested in understanding or replicating the training process, the code is available at the link below.

Training Code URL: https://github.com/YosefLab/scvi-hub-models/blob/main/src/scvi_hub_models/TS_train_all_tissues.ipynb

References

The Tabula Sapiens Consortium. The Tabula Sapiens: A multiple-organ, single-cell transcriptomic atlas of humans. Science, May 2022. doi:10.1126/science.abl4896