【报告摘要】Emerging time-series single-cell RNA sequencing (scRNA-seq) data provide unprecedented opportunities to study dynamic processes of cell populations. However, dynamic inference based on time-series scRNA-seq data is challenging due to the destructive nature of single-cell sequencing. It remains a computational challenge to link the scRNA-seq snapshots sampled at different time points. This requires the development of mathematical models and machine learning methods capable of reconstructing cell population dynamics and the underlying global landscape. We developed a physics-informed neural network framework that combines the Hamilton-Jacobi equation and the neural stochastic differential equation (SDE) to learn cellular dynamics. The proposed physics-informed neural SDE framework not only facilitates accurate prediction of cellular dynamics, but also improves interpretability.

【报告人简介】万林，中国科学院数学与系统科学研究院研究员。2003年获南京大学物理系学士学位，2009年获北京大学数学科学学院博士学位，随后在中国科学院数学与系统科学研究院工作至今。曾先后于美国南加州大学、美国数学生物研究所等机构从事博士后或访问研究。主要研究领域为计算生物学、系统生物学、数据科学和机器学习等。