Speaker: 刘志远，中国科学院动物研究所

Inviter: 万林

Title: GATEVelo——an Expert Model based on Graph Attention Network for deciphering cellular dynamical process

Language: Chinese

Time & Venue: 2024.12.11 16:20-17:00 思源楼415

Abstract: Understanding the dynamic processes underlying cellular development and differentiation is a fundamental goal in biology. RNA velocity analysis has emerged as a powerful tool to infer the future state of individual cells by leveraging the abundances of unspliced and spliced mRNA transcripts in single-cell RNA sequencing (scRNA-seq) data. However, the intrinsic heterogeneity of cells poses significant challenges to existing RNA velocity inference methods. Here, we present a novel RNA velocity model based on a Graph Attention Autoncoder (GATE) that addresses cellular heterogeneity by inferring flexible, cell-gene-specific parameters. Our model introduces a mixture of expert networks, where each expert specializes in a distinct dynamical process associated with specific cell clusters. By integrating these advancements, our model offers a more accurate and detailed understanding of cellular dynamics, with the potential to reveal new insights into developmental processes and gene regulation mechanisms.