Optimal Transport Modeling and Biological Application

← Research overview

We investigate optimal transport and its applications to biological data. Our recent work focuses on modeling cell population dynamics by integrating information from datasets collected across different measurement modalities.

COATI

Cross-Omics Aligned Trajectory Inference

COATI reconstructs continuous cell-state dynamics from time-indexed single-cell snapshots. It uses synchronized optimal transport and neural ODEs to couple trajectory inference across modalities, including gene expression and chromatin accessibility.

A flow learned in the RNA feature space is mapped into the ATAC space through a pretrained cross-modal map. Observations in both spaces constrain the dynamics, while each modality retains its own representation.

Modeling population dynamics across modalities

• Synchronized trajectories. A neural ODE describes cell-state evolution in the primary space, and a fixed cross-modal map projects the trajectories into the secondary space.
• Balanced and unbalanced models. Balanced COATI conserves population mass; unbalanced COATI additionally models growth and depletion.
• Trajectory readouts. The package supports visualization of inferred flows, model-based population changes, and local gene-expression and accessibility readouts along trajectories.

The project documentation includes synthetic and biological examples, installation instructions, and the mathematical formulation.

Related work

• Synchronization of unbalanced dynamical optimal transport across multiple spaces
• Cross-modality trajectory inference integrating scRNA-seq and scATAC-seq
• RAFT-UP: spatial transcriptomics alignment