Trajectory Inference with Growth via Optimal transport and Neural network
TIGON is a dynamic unbalanced optimal transport (OT) model that reconstructs dynamic trajectories and population growth simultaneously as well as the underlying gene regulatory network (GRN) from time-series scRNA-seq snapshots.
The training framework is implemented in PyTorch and Neural ODEs. Given a stable internet connection, it will take several minutes to install these packages:
- pytorch 1.13.1
- scipy 1.10.1
- TorchDiffEqPack 1.0.1
- torchdiffeq 0.2.3
- Recommended: An Nvidia GPU with CUDA support for GPU acceleration
For generating plots to visualize results, the required packages are listed:
- numpy 1.23.5
- seaborn 0.12.2
- matplotlib 3.5.3
$Dataset.npy: data coordinates from different time points. One simulation and three single-cell datasets used for TIGON paper are provided in folder Input/.
--dataset Name of the data set. Options: EMT; Lineage; Bifurcation; Simulation, default = 'Simulation'.
--input_dir Input Files Directory, default='Input/'.
--save_dir Output Files Directory, default='Output/'.
--timepoints Time points of data.
--niters Number of traning iterations.
--lr Learning rate.
--num_samples Number of sampling points per epoch.
--hidden_dim Dimension of hidden layer.
--n_hiddens Number of hidden layers for the neural network learning velocity.
ckpt.pth: save model’s parameters and training errors.
Training process: python3 TIGON.py
Visualization of results: python3 plot_result.py
A Jupyter Notebook of the step-by-step tutorial is accessible from :
Training process: Notebooks/Training.ipynb
Visualization of results: Notebooks/Visualization.ipynb
Single-cell lineage tracing dataset (raw data of Lineage.npy) can be obtained from: Weinreb, Caleb, et al. "Lineage tracing on transcriptional landscapes links state to fate during differentiation." Science 367.6479 (2020): eaaw3381.
TGFB1 induced EMT from A549 cancer cell line dataset (raw data of EMT.npy) can be obtained from: Cook, David P., and Barbara C. Vanderhyden. "Context specificity of the EMT transcriptional response." Nature communications 11.1 (2020): 1-9.
Single-cell qPCR dataset of iPSCs toward cardiomyocytes dataset (raw data of Bifurcation.npy) can be obtained from: Bargaje, Rhishikesh, et al. "Cell population structure prior to bifurcation predicts efficiency of directed differentiation in human induced pluripotent cells." Proceedings of the National Academy of Sciences 114.9 (2017): 2271-2276.
Sha, Y., Qiu, Y., Zhou, P., & Nie, Q. (2023). Reconstructing growth and dynamic trajectories from single-cell transcriptomics data. Nature Machine Intelligence. https://www.nature.com/articles/s42256-023-00763-w#