scMultiNODE: Temporal Single-Cell Data Integration across Unaligned Modalities

scMultiNODE is a model that integrates gene expression and chromatin accessibility measurements in developing single cells while preserving cell type variations and cellular dynamics. scMultiNODE uses autoencoders (AEs) to learn nonlinear low-dimensional cell representation and optimal transport to align cells across different measurements. Next, it utilizes neural ordinary differential equations (ODEs) to explicitly model cell development with a regularization term to learn a dynamic latent space. (bioRxiv preprint)

If you have questions or find any problems with our codes, feel free to submit issues or send emails to [email protected] or other corresponding authors.

(11/01/2024 updates) We have updated major parts of the experiments corresponding to our paper, including scMultiNODE implementation and its integration, integration performance comparison, and downstream analysis.

Requirements

Our codes have been tested in Python 3.7. Required packages are listed in ./installation.

Data

• Raw and preprocessed data of four temporal multi-modal single-cell datasets can be downloaded from here.
• All model integrations and corresponding evaluation metrics on four datasets are available at here (modal_integration.zip).
• Experiment results for downstream analysis (cell path construction and DE genes detection) are available at here (downstream_analysis.zip).
• Data visualization and figures in the paper are available at here (figs.zip).

Models

scMultiNODE is implemented in ./model/dynamic_model.py.

Example Usage

The script of using scMultiNODE for integration is shown in ./modal_integration/Modal_Integration_scMultiNODE.py.

Repository Structure

• data: Scripts for data preprocessing. Some scripts are implemented in R and need installation of Seurat.
• model: Implementation of scMultiNODE model.
• optim: Loss computations, QGW algorithm, evaluation metrics.
• baseline: Implementation of baseline models.
• modal_integration: Run each model on four single-cell datasets and compute integrations. Evaluation metrics computation and comparison.
• downstream_analysis: Use scMultiNODE for cell path construction and finding driver genes.
• hyperparameter_investigation: Ablation study and investigation of hyperparameter settings.
• tuning: Hyperparameter tuning.
• plotting: Integration visualization. Compare model predictions. Paper figures plotting.
• utils: Utility functions.

Bugs & Suggestions

Please report any bugs, problems, suggestions, or requests as a Github issue