Reconstructing developmental trajectories using latent dynamical systems and time-resolved transcriptomics.

Area of Science:

• Developmental Biology
• Computational Biology
• Genomics

Background:

• Single-cell transcriptomics offers snapshots of cellular states, limiting the study of dynamic processes like cell fate decisions.
• Existing methods for temporal inference, such as metabolic labeling and splicing analysis, have inherent limitations.
• Understanding cell fate dynamics is crucial for developmental biology and regenerative medicine.

Purpose of the Study:

• To develop an integrated experimental and computational framework for high-resolution dynamical modeling of cell fate decisions.
• To improve the quality and temporal resolution of single-cell data for studying differentiation.
• To create robust computational tools for inferring and simulating cell trajectories.

Main Methods:

• Developed sci-FATE2, an optimized metabolic labeling technique for enhanced single-cell RNA sequencing data quality.
• Applied VelvetVAE, a variational autoencoder, for accurate gene expression velocity inference.
• Utilized VelvetSDE, a neural stochastic differential equation model, to simulate cell trajectory distributions.

Main Results:

• Profiled 45,000 embryonic stem cells differentiating into neural tube identities using sci-FATE2.
• VelvetVAE demonstrated superior performance in velocity inference compared to existing tools.
• VelvetSDE successfully recapitulated dataset distributions, capturing fate decision boundaries and gene expression dynamics.

Conclusions:

• The developed framework transforms single-cell analyses from static descriptions to dynamic models of biological processes.
• This approach provides powerful tools for investigating the complex mechanisms underlying developmental cell fate decisions.
• The integration of improved experimental methods and advanced computational modeling opens new avenues for studying cellular dynamics.