TemporalVAE

Temporal mapping of single cells from time-series atlas with time-predicting VAE

Reproducibility

For reproducibility of the manuscript's analyses, the scripts for generating figures are available at TemporalVAE-reproducibility folder/submodule.

Installation

Quick install can be achieved via pip (python >=3.8)

Step 0: create a conda environment and activate it:

conda create -n tvae python=3.10.9
conda activate tvae

# Optional: add jupyter lab kernal
pip install ipykernel
python -m ipykernel install --user --name tvae --display-name "tvae"

Step 1: install TemporalVAE from GitHub:

# or developing version
pip install -U git+https://github.com/StatBiomed/TemporalVAE-release

# for published version (TO AVAILABLE SOON)
# pip install -U TemporalVAE

Quick Usage

Reference examples can be found at examples folder, including

• training: hEmbryo8_training.ipynb

• cross-validation: hEmbryo8_Xiang_CV.ipynb

• predicting: [TO BE ADDED]

Future plan for easier use

Here are the future plan for easier use (TO IMPLEMENT):

1. Import TemporalVAE and create an object of the class TVAE.

import TemporalVAE as tvae

tvae_model = tvae.TVAE()
tvae_model.fit(X_atlas, t_atlas)

# predict query or training data
Z_query, y_query = tvae_model.predict(X_query)
Z_atlas, y_atlas = tvae_model.predict(X_atlas)

2. Map to the same UMAP as the reference data

import UMAP

umap_model = UMAP.umap()
umap_model.fit(Z_atlas)

atlas_umap = umap_model.transform(Z_atlas)
query_umap = umap_model.transform(Z_query)

Reference

Y. Liu et al. Atlas-assisted temporal mapping of time-series single-cell transcriptomes during embryogenesis. (to appear)