Efficient differential expression analysis of large-scale single-cell transcriptomics data using Dreamlet
Gabriel E. Hoffman, Donghoon Lee, Jaroslav M. Bendl, N. M. Prashant, Aram Hong, Clara Casey et al. · 2026 · 65 citationsRead the paper
Advances in single-cell and -nucleus transcriptomics have enabled generation of increasingly large-scale datasets from hundreds of subjects and millions of cells. These studies promise to give unprecedented insight into the cell type specific biology of human disease. Yet performing differential expression analyses across subjects remains difficult due to challenges in statistical modeling of these complex studies and scaling analyses to large datasets. Our open-source R package dreamlet (DiseaseNeurogenomics.github.io/dreamlet) uses a pseudobulk approach based on precision-weighted linear mixed models to identify genes differentially expressed with traits across subjects for each cell cluster. Designed for data from large cohorts, dreamlet is substantially faster and uses less memory than existing workflows, while supporting complex statistical models and controlling the false positive rate. We demonstrate computational and statistical performance on published datasets, and a novel dataset of 1.4M single nuclei from postmortem brains of 150 Alzheimer's disease cases and 149 controls.
What the full text shows
Working prototype- Dreamlet processes pseudobulk for 1000 donors across 8 cell types (4.1M cells) significantly faster and with lower memory than competing methods. — 31 min, 26 Gb memory (Single-threaded execution on a compute cluster)
- Dreamlet fits precision-weighted linear mixed models for 12 cell types across 326 subjects. — 15 min wall time (45 CPU minutes) (Parallelized across multiple CPU cores)
- Dreamlet identifies a high-effect upregulation of PTPRG specifically in microglia of Alzheimer's disease cases. — log2 fold change of 1.59, p-value 2.94e-28 (1.4M single nuclei from 150 AD cases and 149 controls)
- In classical monocytes of COVID-19 patients, expression variation across disease states was the strongest source of variation. — median of 11.4% variance (674K cells from 110 donors)
- Dreamlet controls the false positive rate while retaining power in the presence of high-dimensional batch effects. (Compared to fixed effects models and GLMMs (glmer, MAST))
- Setup
- Datasets include 1.4M single nuclei (299 AD/control donors), 674K cells (110 COVID-19/sepsis/control donors), and prostate cancer bone metastasis samples.
- How
- Dreamlet uses a pseudobulk approach combining precision-weighted linear mixed models with a two-stage weighting process (Poisson initialization followed by empirical mean-variance trend fitting). It utilizes the H5AD format for on-disk storage and parallel processing to scale to millions of cells.
- Limitations
- Simulating single-cell transcriptomics data that accurately recapitulates real-world complexity is challenging. · Technical batch effects can be substantial for some genes. · Statistical power for each cell cluster varies based on the number of nuclei observed per subject.
Summarized by Gemma 4 from the open-access full text · 14 Oct 2026. Read it