Learning Latent Trajectories in Developmental Time Series with Hidden-Markov Optimal Transport.

Area of Science:

• Developmental Biology
• Computational Biology
• Genomics

Background:

• Understanding cell differentiation sequences is crucial for developmental biology.
• Single-cell and spatial sequencing provide data for developmental studies.
• Inferring differentiation maps requires trajectory analysis and cell type coarse-graining.

Purpose of the Study:

• To introduce Hidden-Markov Optimal Transport (HM-OT), an algorithm for learning cell type transitions during development.
• To simultaneously group cells into types and learn differentiation pathways from time-series transcriptomics data.
• To provide a method for unsupervised or semi-supervised learning of differentiation maps.

Main Methods:

• HM-OT utilizes low-rank optimal transport to align time-series samples.
• The algorithm learns clusterings and differentiation maps by minimizing total transport cost.
• It assumes a Markov property for latent cell-type trajectories.

Main Results:

• HM-OT successfully identifies cell types and their differentiation transitions.
• The algorithm was validated on zebrafish development data (Stereo-seq).
• HM-OT demonstrated scalability on large-scale mouse embryonic development data (Stereo-seq).

Conclusions:

• HM-OT offers a robust method for reconstructing developmental trajectories and differentiation maps.
• The algorithm can be applied in unsupervised or semi-supervised settings.
• HM-OT advances the analysis of single-cell transcriptomics for developmental studies.