Computational Cell Biology Faculty Candidate Research Seminar – Dongyuan Song

Speaker: Dongyuan Song – PhD candidate, University of California, Los Angeles

This seminar will be presented in a hybrid format. The speaker will be delivering their talk in person at the MSL Lecture Theatre (room 102). Audience members are welcome to attend either in person or via the zoom link. Those connecting via zoom will be able to ask questions during the Q&A portion using the chat function.

Zoom link: https://ubc.zoom.us/meeting/register/u5wpf-2rrDIoE9JhILZP49JCJwWMw0Ur6N77

Talk title: Promoting statistical rigor in single-cell omics: Generative models and in silico control data

Abstract: In the era of single-cell biology, new computational tools have been continuously developed to address diverse analytical tasks, but their statistical interpretations (e.g., p-values) are often questionable. In this talk, I will first introduce scDesign3, a unified generative model for single-cell and spatial multi-omics. By simulating realistic single-cell data, scDesign3 can help researchers choose the best computational tool and guide the experimental design. Next, I will illustrate how this generative model can be further used to perform more rigorous statistical tests in single-cell clustering analysis. By generating the in silico control data, we successfully identify spurious clusters and reduce the false discovery of “fake” cell type markers.