Joint trajectory inference for single-cell genomics using deep learning with a mixture prior.

Area of Science:

• Computational Biology
• Genomics
• Developmental Biology

Background:

• Single-cell sequencing data analysis relies on trajectory inference to understand cell differentiation and development.
• Existing trajectory inference tools often lack robust statistical models and uncertainty quantification.

Purpose of the Study:

• To introduce VITAE (Variational Inference for Trajectory by AutoEncoder), a novel statistical framework for robust cell trajectory inference.
• To enhance interpretability and computational efficiency in trajectory analysis through a hierarchical mixture model and variational autoencoders.

Main Methods:

• Developed VITAE, integrating a latent hierarchical mixture model with variational autoencoders for trajectory inference.
• Enabled simultaneous trajectory inference and data integration to handle biological and technical heterogeneity.
• Utilized posterior approximations for uncertainty quantification of cell projections.

Main Results:

• VITAE demonstrated superior performance compared to state-of-the-art methods on synthetic and real datasets with diverse topologies.
• Successfully applied VITAE to jointly analyze mouse neocortex single-cell RNA sequencing data, revealing detailed projection neuron lineages.
• Showcased VITAE's effectiveness in reducing batch effects and uncovering finer cellular structures across datasets.
• Validated VITAE's utility in integrative multiomic analyses of continuous cell populations.

Conclusions:

• VITAE provides a statistically rigorous and computationally efficient approach for cell trajectory inference and data integration.
• The method enhances the understanding of developmental processes and lineage hierarchies, particularly in complex biological systems.
• VITAE is effective for analyzing single-cell and multiomic data, offering improved accuracy and reduced batch effects.