OTMODE: Optimal Transport-Based Framework for Differential Feature Identification in Single-Cell Multi-Omics

OTMODE is a computational framework built on Optimal Transport (OT) theory for improving cell-type annotation accuracy and for identifying differential features across conditions in single-cell multi-omics data.

It provides interpretable metrics, including Sinkhorn distances, to compare cell types and clusters across predicted and true annotations. Please see our architecture of the OTMODE framework for more details!

To optimize the performance of OTMODE in differential feature detection, we evaluated its behavior under various parameter settings. Our results showed that parameter values between 0.1 and 1.0 generally yielded the highest performance.

🔬 Built with Python

We provide a simple installation process to help users quickly set up and start using OTMODE for multi-omics single-cell data analysis.

You can install OTMODE using pip:

pip install otmode

Alternatively, to install the latest development version from GitHub:

git clone https://github.com/your-username/otmode.git
cd otmode
pip install -e .

Make sure you have Python 3.8+ and the required dependencies listed in requirements.txt.

We provide interactive tutorials to help users get started with applying OTMODE to multi-omics single-cell data. These tutorials demonstrate real use cases such as cell type annotation and differential feature detection.

All tutorials are provided as Jupyter notebooks and are located in the tutorials/ folder.

Each tutorial is fully commented and includes the following steps:

• 📂 Data Loading and Preparation
  Load example or user-provided multi-omics data and prepare it for analysis.

• ⚙️ Application of the OTMODE Algorithm
  Apply the OTMODE method to integrate modalities and infer latent representations.

• 📊 Visualization of Results
  Generate plots such as UMAP, heatmaps, and feature importance to interpret results.

• 🔍 Interpretation of Outputs
  Understand clustering, annotations, or differential features identified by OTMODE.

✅ Recommended Environment

• Python ≥ 3.8
• Libraries: scanpy, anndata, numpy, matplotlib, seaborn, and others listed in requirements.txt