stVCR: spatiotemporal dynamics of single cells

Area of Science:

• Developmental Biology
• Computational Biology
• Genomics

Background:

• Time-series spatial transcriptomics offers insights into cell behavior over time.
• Reconstructing spatiotemporal dynamics is challenging due to destructive sequencing and differing coordinate systems across replicates.
• Inferring cell migration from static snapshots is particularly difficult.

Purpose of the Study:

• To develop a novel deep learning framework, spatiotemporal video cassette recorder (stVCR), for reconstructing continuous cell dynamics.
• To enable end-to-end analysis of cell differentiation, proliferation, migration, and spatial alignment.
• To provide interpretable insights into the interplay between cell phenotype transitions and spatial dynamics.

Main Methods:

• Developed stVCR, a generative deep learning framework integrating dynamical optimal transport and density matching.
• Incorporated biologically informed priors to maintain spatial structure during reconstruction.
• Utilized unbalanced optimal transport and rigid transformation-invariant density matching for robust analysis.

Main Results:

• stVCR successfully reconstructs continuous cell differentiation, proliferation, and migration in physical space.
• The framework allows for interpretable analysis of phenotype transitions and spatial migration.
• Demonstrated effectiveness and robustness using both simulated and real biological datasets.

Conclusions:

• stVCR provides a powerful tool for analyzing complex spatiotemporal dynamics in biological systems.
• The framework can uncover intricate cell behaviors during processes like regeneration and development.
• Applied stVCR to reveal novel spatiotemporal insights in axolotl brain regeneration and Drosophila embryo development.