单细胞识别DEGs还是Pseudobulk方法更可靠？

image.png

article: 10.1038/s41467-021-25960-2, Nature Communications, 2021

Main content

1. establish the methodological framework -> to curate a resource of ground truth datasets
   the matched bulk RNA and scRNA data with the same perturbation and sequenced in the same laboaratories.
2. The most popular methods that were benchmarked in this paper (grey bar).

image.png

3. Evaluations

• concordance between DE results in bulk versus scRNA-seq datasets using concordance curve (AUCC)
• concordance between GO results using the DEGs.
• split the control data into two samples randomly for test

4. Results

• the pseudobulk methods outperform the specialized sc-DE methods, alse in the proteomics data, because they can account for the the intrinsic variability of biological replicates to generate the biologically accurate results.
• sc-DE methods tend to identify the highly expressed genes as the DEGs, even they remain unchanged.
• negative binomial generalized linear mixed models can generate accurate results but are time-consuming.