Multi-omic single-cell velocity models epigenome-transcriptome interactions and improves cell fate prediction

Area of Science:

• Single-cell multi-omics
• Computational biology
• Epigenetics and transcriptomics

Background:

• Understanding temporal dynamics of gene regulation requires integrating multiple molecular layers within single cells.
• Current methods often analyze epigenomic and transcriptomic data separately, limiting insights into their interplay.
• RNA velocity provides a framework for inferring future gene expression states but typically lacks epigenetic information.

Purpose of the Study:

• To develop a computational model, MultiVelo, that integrates multi-omic single-cell data to infer temporal relationships between epigenome and transcriptome.
• To enhance the accuracy of cell fate prediction by incorporating epigenomic dynamics into gene expression modeling.
• To characterize gene regulatory dynamics and cell state transitions using integrated multi-omic data.

Main Methods:

• Developed MultiVelo, a differential equation model extending the RNA velocity framework to incorporate epigenomic data (chromatin accessibility).
• Utilized a probabilistic latent variable model within MultiVelo to estimate switching times and rates for chromatin accessibility and gene expression.
• Applied MultiVelo to multi-omic single-cell datasets from various tissues (brain, skin, blood).

Main Results:

• MultiVelo accurately predicts cell fates by integrating epigenomic and transcriptomic data, outperforming RNA-only velocity estimates.
• Identified two distinct classes of genes based on the temporal order of chromatin closing and transcription cessation.
• Discovered four types of cell states, including coupled and decoupled epigenome-transcriptome dynamics, and revealed time lags between transcription factor activity and gene expression.
• Uncovered temporal relationships between disease-associated single nucleotide polymorphism (SNP) accessibility and linked gene expression.

Conclusions:

• MultiVelo provides a powerful framework for analyzing temporal dynamics in single-cell multi-omic data.
• The model reveals novel insights into gene regulatory mechanisms and cell state transitions by integrating epigenomic and transcriptomic information.
• This approach advances our understanding of how epigenetic changes influence gene expression over time and their role in cellular processes and disease.