TotalVI is a variational inference model for single-cell RNA-seq as well as protein data that can learn an underlying latent space, integrate technical batches, impute dropouts, and predict protein expression given gene expression or missing protein data given gene expression and protein data for a subset of proteins. The learned low-dimensional latent representation of the data can be used for visualization and clustering.

TotalVI takes as input a scRNA-seq gene expression and protein expression matrix with cells and genes. We provide an extensive user guide.

See our original manuscript for further details of the model: TotalVI 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

Bone marrow CITE-seq data generated for NEURIPS 2021.

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:

Modality: rna

Metric	Training Value	Validation Value
Mean Absolute Error	1.32	1.31
Pearson Correlation	0.97	0.96
Spearman Correlation	0.90	0.90
R² (R-Squared)	0.89	0.88

Modality: protein

Metric	Training Value	Validation Value
Mean Absolute Error	1.46	1.42
Pearson Correlation	0.47	0.48
Spearman Correlation	0.69	0.69
R² (R-Squared)	-0.62	-0.62

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:

Modality: rna

Metric	Training Value
Mean Absolute Error	3.33
Pearson Correlation	0.74
Spearman Correlation	0.88
R² (R-Squared)	-0.93

Modality: protein

Metric	Training Value
Mean Absolute Error	1.55
Pearson Correlation	0.83
Spearman Correlation	0.84
R² (R-Squared)	-1.43

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:

Modality: rna

Index	gene_f1	lfc_mae	lfc_pearson	lfc_spearman	roc_auc	pr_auc	n_cells
CD14+ Mono	0.96	0.17	0.99	1.00	0.13	0.02	21693.00
CD4+ T activated	0.96	0.35	0.84	0.99	0.09	0.02	6966.00
CD4+ T naive	0.97	0.55	0.69	0.98	0.10	0.02	5897.00
NK	0.94	0.44	0.76	0.98	0.07	0.02	5434.00
Reticulocyte	0.96	1.95	0.60	0.80	0.33	0.03	4272.00
Erythroblast	0.96	0.54	0.86	0.99	0.20	0.03	4039.00
Naive CD20+ B IGKC+	0.98	0.68	0.67	0.98	0.07	0.02	3990.00
CD8+ T naive	0.95	0.67	0.64	0.98	0.07	0.02	3107.00
CD16+ Mono	0.91	0.76	0.66	0.98	0.05	0.02	2635.00
NK CD158e1+	0.95	1.19	0.61	0.95	0.08	0.02	2167.00
Naive CD20+ B IGKC-	0.96	1.38	0.60	0.96	0.10	0.02	1979.00
G/M prog	0.88	0.36	0.80	0.98	0.01	0.01	1881.00
pDC	0.92	0.92	0.68	0.98	0.05	0.02	1758.00
HSC	0.91	0.42	0.75	0.99	0.03	0.02	1703.00
cDC2	0.88	0.36	0.79	0.98	0.01	0.02	1702.00
Lymph prog	0.93	0.49	0.78	0.99	0.04	0.02	1681.00
Transitional B	0.96	1.40	0.64	0.96	0.08	0.02	1575.00
Proerythroblast	0.87	0.43	0.85	0.99	0.07	0.02	1512.00
CD8+ T CD57+ CD45RO+	0.93	1.82	0.61	0.95	0.10	0.02	1470.00
Normoblast	0.96	2.64	0.63	0.85	0.32	0.03	1435.00
CD8+ T CD57+ CD45RA+	0.89	1.55	0.57	0.93	0.07	0.02	1303.00
CD8+ T TIGIT+ CD45RO+	0.91	1.80	0.56	0.91	0.09	0.02	1160.00
CD4+ T activated integrinB7+	0.92	2.05	0.53	0.92	0.13	0.02	1056.00
CD8+ T TIGIT+ CD45RA+	0.92	2.34	0.53	0.90	0.16	0.02	1032.00
CD8+ T CD49f+	0.92	1.83	0.56	0.92	0.09	0.02	912.00
CD8+ T CD69+ CD45RO+	0.90	2.42	0.57	0.90	0.14	0.02	897.00
B1 B IGKC+	0.89	2.12	0.62	0.92	0.15	0.01	820.00
MAIT	0.90	2.18	0.57	0.91	0.13	0.02	756.00
CD8+ T CD69+ CD45RA+	0.91	2.66	0.56	0.89	0.19	0.02	740.00
MK/E prog	0.90	0.80	0.67	0.98	0.04	0.02	690.00
gdT CD158b+	0.89	2.06	0.58	0.90	0.12	0.02	674.00
B1 B IGKC-	0.88	2.31	0.63	0.91	0.16	0.02	613.00
T reg	0.87	2.33	0.55	0.87	0.12	0.01	609.00
ILC1	0.89	2.68	0.51	0.85	0.18	0.02	552.00
Plasma cell IGKC+	0.82	2.41	0.44	0.82	0.19	0.01	288.00
Plasma cell IGKC-	0.88	3.00	0.50	0.81	0.20	0.01	239.00
ILC	0.87	4.29	0.59	0.73	0.28	0.02	238.00
Plasmablast IGKC+	0.84	3.16	0.53	0.84	0.21	0.01	232.00
gdT TCRVD2+	0.84	3.82	0.56	0.75	0.22	0.01	191.00
Plasmablast IGKC-	0.83	4.08	0.54	0.78	0.24	0.02	130.00
CD4+ T CD314+ CD45RA+	0.81	4.40	0.62	0.73	0.22	0.01	93.00
dnT	0.73	5.84	0.53	0.52	0.35	0.02	56.00
CD8+ T naive CD127+ CD26- CD101-	0.77	5.38	0.57	0.60	0.29	0.02	42.00
T prog cycling	0.65	4.82	0.65	0.73	0.26	0.02	24.00
cDC1	0.65	5.83	0.58	0.57	0.28	0.02	18.00

Modality: protein

Index	gene_f1	lfc_mae	lfc_pearson	lfc_spearman	roc_auc	pr_auc	n_cells
CD14+ Mono	0.92	0.38	0.99	0.99	0.24	0.14	21693.00
CD4+ T activated	0.92	0.23	0.99	0.98	0.24	0.14	6966.00
CD4+ T naive	0.92	0.19	0.99	0.97	0.12	0.13	5897.00
NK	0.92	0.21	0.98	0.98	0.17	0.14	5434.00
Reticulocyte	0.46	0.91	0.95	0.98	0.29	0.16	4272.00
Erythroblast	0.62	0.97	0.96	0.98	0.27	0.15	4039.00
Naive CD20+ B IGKC+	0.92	0.31	0.99	0.99	0.36	0.13	3990.00
CD8+ T naive	0.92	0.24	0.98	0.98	0.64	0.24	3107.00
CD16+ Mono	0.92	0.24	0.99	0.99	0.76	0.32	2635.00
NK CD158e1+	0.92	0.28	0.98	0.98	0.74	0.44	2167.00
Naive CD20+ B IGKC-	0.92	0.35	0.99	0.99	0.79	0.38	1979.00
G/M prog	0.92	0.33	0.98	0.96	0.92	0.92	1881.00
pDC	1.00	0.26	0.98	0.96	0.92	0.93	1758.00
HSC	0.85	0.35	0.97	0.95	0.54	0.57	1703.00
cDC2	0.92	0.31	0.98	0.98	1.00	0.95	1702.00
Lymph prog	0.92	0.34	0.97	0.98	0.85	0.85	1681.00
Transitional B	0.85	0.54	0.97	0.98	0.77	0.79	1575.00
Proerythroblast	0.77	0.69	0.94	0.97	0.77	0.79	1512.00
CD8+ T CD57+ CD45RO+	0.92	0.49	0.97	0.97	0.94	0.52	1470.00
Normoblast	0.54	1.01	0.95	0.97	0.78	0.64	1435.00
CD8+ T CD57+ CD45RA+	0.85	0.31	0.98	0.98	0.99	0.96	1303.00
CD8+ T TIGIT+ CD45RO+	0.92	0.25	0.99	0.97	0.85	0.86	1160.00
CD4+ T activated integrinB7+	0.92	0.30	0.98	0.96	0.99	0.96	1056.00
CD8+ T TIGIT+ CD45RA+	0.92	0.25	0.99	0.97	1.00	0.99	1032.00
CD8+ T CD49f+	0.92	0.22	0.98	0.96	0.99	0.96	912.00
CD8+ T CD69+ CD45RO+	0.92	0.25	0.98	0.98	1.00	0.99	897.00
B1 B IGKC+	0.92	0.34	0.98	0.97	1.00	0.99	820.00
MAIT	0.92	0.28	0.98	0.98	0.99	0.94	756.00
CD8+ T CD69+ CD45RA+	1.00	0.23	0.98	0.98	1.00	1.00	740.00
MK/E prog	0.85	0.26	0.96	0.96	0.53	0.46	690.00
gdT CD158b+	0.85	0.36	0.96	0.97	0.99	0.92	674.00
B1 B IGKC-	1.00	0.37	0.98	0.97	1.00	1.00	613.00
T reg	0.77	0.26	0.98	0.97	0.99	0.93	609.00
ILC1	0.92	0.23	0.98	0.96	1.00	0.98	552.00
Plasma cell IGKC+	0.77	0.31	0.96	0.96	0.91	0.85	288.00
Plasma cell IGKC-	0.92	0.25	0.96	0.97	0.92	0.92	239.00
ILC	0.77	0.35	0.97	0.96	0.99	0.92	238.00
Plasmablast IGKC+	0.92	0.25	0.96	0.97	0.77	0.74	232.00
gdT TCRVD2+	0.85	0.32	0.94	0.91	0.99	0.92	191.00
Plasmablast IGKC-	0.92	0.28	0.95	0.93	0.85	0.85	130.00
CD4+ T CD314+ CD45RA+	0.69	0.77	0.94	0.93	0.98	0.86	93.00
dnT	0.92	0.42	0.90	0.87	0.68	0.55	56.00
CD8+ T naive CD127+ CD26- CD101-	0.69	0.55	0.92	0.88	0.98	0.80	42.00
T prog cycling	0.69	0.63	0.86	0.79	0.51	0.60	24.00
cDC1	0.38	0.64	0.80	0.74	0.59	0.55	18.00

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_latent": 20,
    "gene_dispersion": "gene",
    "protein_dispersion": "protein",
    "gene_likelihood": "nb",
    "latent_distribution": "normal",
    "empirical_protein_background_prior": null,
    "override_missing_proteins": false
}

Setup Data Arguments

Arguments passed to setup_anndata of the original model:

{
    "rna_layer": "counts",
    "protein_layer": "counts",
    "batch_key": "batch",
    "size_factor_key": null,
    "categorical_covariate_keys": null,
    "continuous_covariate_keys": null,
    "modalities": {
        "rna_layer": "rna",
        "protein_layer": "protein",
        "batch_key": "rna"
    }
}

Data Registry

Registry elements for AnnData manager:

Registry Key	scvi-tools Location
X	adata.mod['rna'].layers['counts']
batch	adata.mod['rna'].obs['_scvi_batch']
labels	adata.obs['_scvi_labels']
latent_qzm	adata.obsm['totalvi_latent_qzm']
latent_qzv	adata.obsm['totalvi_latent_qzv']
minify_type	adata.uns['_scvi_adata_minify_type']
observed_lib_size	adata.obs['observed_lib_size']
proteins	adata.mod['protein'].layers['counts']

Data is Minified: False

Summary Statistics

Summary Stat Key	Value
n_batch	12
n_cells	90261
n_extra_categorical_covs	0
n_extra_continuous_covs	0
n_labels	1
n_latent_qzm	20
n_latent_qzv	20
n_proteins	134
n_vars	4000

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/

References

A sandbox for prediction and integration of DNA, RNA, and proteins in single cells; https://datasets-benchmarks-proceedings.neurips.cc/paper_files/paper/2021/file/158f3069a435b314a80bdcb024f8e422-Paper-round2.pdf.

scvi-tools
/

bone_marrow_cite_totalvi

Model Description

Metrics

Model Properties

References

Collection including scvi-tools/bone_marrow_cite_totalvi

TotalVI