报告题目：On the mathematics of RNA velocity

报告人：周沛劼 研究员（北京大学）

报告时间：6月17日11:00-12:00

报告地点：数理楼245

报告摘要：RNA velocity has emerged as a popular approach for studying cell stemness and lineage during development from scRNA-seq data. Here we provide a mathematical analysis of the RNA velocity model, spanning from its dynamical systems theory to computational algorithm design. Theoretically, we have derived the analytical solutions of the RNA velocity model from stochastic perspectives, and present a parameter inference framework on maximum likelihood estimation. For computational implementation, we address the uncertainty quantification strategy for parameters inferred via the EM algorithm, and establish a rational time-scale fixation method for determining a global gene-shared latent time across cells. Moreover, we are motivated by the dynamical systems theory to explore the continuum limits of velocity-induced dynamics, which are designed for downstream analysis of streamline visualization and root finding, and then develop an optimal criterion for selecting velocity kernel bandwidth in relation to sample size based on the theoretical study. Overall, our mathematical study contributes to enhance the understanding and application of RNA velocity models.

报告人简介：周沛劼，北京大学前沿交叉学科研究院国际机器学习中心研究员，博士生导师。入选国家海外高层次青年人才计划。2019年毕业于北京大学数学科学学院，2020-2023年担任美国加州大学尔湾分校访问助理教授。研究领域为计算系统生物学，主要科研兴趣为数据驱动的AI动力学建模与计算，研究成果发表在Nature Methods, Nature Communications, Physical Review X, Molecular Systems Biology, Nature Genetics,Nature Machine Intelligence,Nucleic Acids Research, Briefings in Bioinformatics等交叉学科期刊。