TopoVelo stands for Topological velocity inference. It is a tool for jointly modeling temporal gene expression and spatial cellular dynamics from spatial transcriptomic data. The method applies a graph variational autoencoder to recover spatially-coupled RNA velocity and reveal the spatial migration of cells during tissue growth.
To install the TopoVelo package, you can use pip:
git clone https://github.com/welch-lab/TopoVelo.git
cd TopoVelo
pip install .
The package provides function modules for training and evaluating a TopoVelo model and plotting results. The default pipeline with minimal hyperparameter tuning is shown as follows:
import scanpy as sc
import scvelo as scv
import topovelo as tpv
import torch
# Load .h5ad to AnnData
adata = sc.read(< path to .h5ad >)
# Preprocessing
spatial_key = 'X_spatial'
tpv.preprocess(adata, n_gene=200, spatial_key=spatial_key)
tpv.build_spatial_graph(adata, spatial_key)
# Create a VAE object
device = torch.device('cpu') if torch.cuda.is_available() else torch.device('cpu')
vae = tpv.VAE(adata, tmax=20, dim_z=5, device=device)
# Training
vae.train(adata, adata.obsp['spatial_graph'], spatial_key)
# Save results
vae.save_model(< path to model parameters >, 'encoder', 'decoder')
vae.save_anndata(adata, 'gat', < path to output >, file_name="adata_out.h5ad")
Examples of advanced usage, model evaluation and plotting can be found in the tutorial.
GNU General Public License v3.0
The package is currently maintained by Yichen Gu. Please contact the maintainer via [email protected].