Single Cell Seminar: Liang He, Duke University

Liang He: NEBULA: a fast negative binomial mixed model for differential analysis of large-scale multi-subject single-cell data (Single Cell Seminars)

Abstract:
Large-scale multi-subject single-cell data become increasingly available in recent years. Accurate and efficient differential analysis plays a pivotal role in identifying marker genes, detecting biologically relevant effects, performing co-expression analysis, and generating normalized data for downstream analysis including clustering. Negative binomial mixed models (NBMMs), accounting for both cell-level and subject-level overdispersions, are an ideal model for handling such an intrinsic hierarchical structure. However, NBMMs are very computationally demanding. In this talk, we introduce an efficient NEgative Binomial mixed model Using a Large-sample Approximation (NEBULA). NEBULA is orders of magnitude faster than existing tools and controls false-positive errors in marker gene identification and co-expression analysis. The speed gain is achieved by analytically solving high-dimensional integrals instead of using the Laplace approximation. Using NEBULA in Alzheimer’s disease cohort data sets, we found that the cell-level expression of APOE correlated with that of other genetic risk factors (including CLU, CST3, TREM2, C1q, and ITM2B) in a cell-type-specific pattern and an isoform-dependent manner in microglia. NEBULA opens up a new avenue for the broad application of mixed models to large-scale multi-subject single-cell data.

Join Zoom Meeting
https://tu-dresden.zoom.us/j/89417259251?pwd=d2FqVk1GWFhQeE1hWmNYelc5a1ZhZz09

Meeting ID: 894 1725 9251
Passcode: .?G*1CWn