|
Abstract:
|
Since its introduction, RNA-seq has generated rich data and have shed insights on many tissue-specific disease mechanisms, including brain. However, most of RNA-seq data are collected from bulk tissue samples, and they represent a heterogeneous mixture of different cell types. The deconvolution of bulk RNA-seq data into cell type-specific characterizations has been an important topic for researchers. Yet the multiple cell types and heterogeneity between brain regions add to difficulty of correctly estimating proportions of cell types. With the increasing applications of single-cell/single-nuclei RNA-seq technologies, cell type-specific expression profiles can be characterized with more accuracy and thus makes deconvolution of bulk RNA-seq data possible. Here I present a new method of deconvolution of bulk RNA-seq with snRNA data using a regularized model of weighted least squares regression. Results show that our method can outperform some commonly used models, such as MuSiC and CIBERSORT, in estimating the proportions of brain cell types. With the cell type-specific information, helpful insights can be gained from the large volume of RNA-seq datasets.
|
