InterVelo: a mutually enhancing model for estimating pseudotime and RNA velocity in multi-omic single-cell data

Area of Science:

• Computational Biology
• Single-cell Genomics
• Machine Learning in Biology

Background:

• RNA velocity analysis is crucial for understanding cellular dynamics from snapshot single-cell data.
• Existing RNA velocity methods often rely on simplifying assumptions, potentially introducing biases.
• A novel deep learning framework, InterVelo, is introduced to address these limitations.

Purpose of the Study:

• To develop a deep learning framework, InterVelo, for simultaneous estimation of cellular pseudotime and RNA velocity.
• To improve the accuracy and robustness of RNA velocity and pseudotime inference compared to existing methods.
• To enable more reliable analysis of transcriptional dynamics and gene activity.

Main Methods:

• InterVelo employs an unsupervised approach where cellular time guides RNA velocity estimation.
• RNA velocity estimations are used iteratively to refine the pseudotime trajectory.
• The framework is benchmarked against existing methods using simulated and real single-cell datasets.

Main Results:

• InterVelo demonstrates superior performance in recovering both pseudotime and RNA velocity.
• The method provides more precise velocity estimations in direction and magnitude across diverse scenarios.
• InterVelo successfully identifies driver genes and facilitates gene activity enrichment analysis, with potential for multi-omic data integration.

Conclusions:

• InterVelo offers a robust and accurate deep learning framework for RNA velocity and pseudotime inference.
• The framework enhances the analysis of transcriptional dynamics and gene regulatory networks.
• InterVelo's flexibility supports integration with multi-omic data for comprehensive biological system analysis.