Inferring single-cell transcriptomic dynamics with structured latent gene expression dynamics.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) data analysis relies on gene expression dynamics for inferring cellular trajectories.
• Traditional RNA velocity methods use strict assumptions that can fail with complex biological systems, such as multiple distinct cell lineages or time-dependent molecular rates.
• Accurate trajectory inference is crucial for understanding developmental processes and cellular differentiation.

Purpose of the Study:

• To develop a novel deep learning framework, LatentVelo, for robust gene expression dynamics modeling in scRNA-seq data.
• To overcome limitations of traditional RNA velocity methods by accommodating complex lineage structures and variable molecular rates.
• To provide a more accurate and comprehensive method for inferring cellular developmental trajectories and regulatory states.

Main Methods:

• LatentVelo employs a variational autoencoder to embed cells into a low-dimensional latent space, capturing essential gene dynamics.
• Neural ordinary differential equations are utilized to model differentiation dynamics within this latent space.
• The framework infers a latent regulatory state to manage individual cell dynamics and model multiple lineages.

Main Results:

• LatentVelo successfully infers latent trajectories, representing the inferred developmental paths of individual cells.
• The method effectively models complex scenarios with multiple lineages and time-dependent gene expression dynamics.
• The dynamics-based embedding in LatentVelo demonstrates superior batch correction capabilities compared to standard autoencoder methods.

Conclusions:

• LatentVelo offers a powerful deep learning approach for analyzing gene expression dynamics in scRNA-seq data.
• The inferred latent trajectories and regulatory states provide deeper insights into cellular differentiation and development.
• LatentVelo advances scRNA-seq trajectory inference by offering improved accuracy, robustness, and batch correction.