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

TL;DR

InterVelo is a new deep learning method that improves the estimation of pseudotime and RNA velocity in single-cell data by integrating them in a mutually enhancing way.

Contribution

InterVelo introduces a novel framework that simultaneously learns pseudotime and RNA velocity, enhancing accuracy and robustness.

Findings

InterVelo outperforms existing methods in recovering pseudotime and RNA velocity across diverse datasets.

The model provides more accurate velocity estimations in terms of direction and magnitude.

InterVelo successfully identifies driver genes and supports gene activity enrichment analysis.

Abstract

RNA velocity has become a powerful tool for uncovering transcriptional dynamics in snapshot single-cell data. However, current RNA velocity approaches often assume constant transcriptional rates and treat genes independently with gene-specific times, which may introduce biases and deviate from biological realities. Here, we present InterVelo, a novel deep learning framework that simultaneously learns cellular pseudotime and RNA velocity.

InterVelo leverages an unsupervised cellular time to guide RNA velocity estimation, while the estimated RNA velocity in turn refines the direction of pseudotime. By benchmarking InterVelo against existing methods on both simulated and real datasets, we demonstrate its superior performance in recovering pseudotime and RNA velocity. InterVelo yields more precise velocity estimations in terms of both direction and magnitude, with outstanding robustness across diverse scenarios. Furthermore, it successfully identifies driver genes and enables reliable gene activity enrichment analysis. The flexible architecture of InterVelo also allows for the integration of multi-omic data, enhancing its applicability to complex biological systems.

InterVelo is implemented using Python, and the code is available on GitHub https://github.com/yurouwang-rosie/InterVelo and has been archived with a DOI https://doi.org/10.5281/zenodo.16158798 for reproducibility.

Figures