# Efficient differential expression analysis of large-scale single-cell transcriptomics data using Dreamlet

- **DOI:** [10.1038/s41467-026-75680-8](https://doi.org/10.1038/s41467-026-75680-8)
- **Rating:** Strong (4.0 out of 5)
- **Verdict:** Dreamlet advances scalable donor-level single-cell analysis with calibrated mixed models, but its broad superiority claims exceed the benchmark evidence.
- **Framework version:** v5.13
- **Pipeline version:** v0.4
- **Evaluation date:** 2026-11-22
- **Evaluation page:** https://www.nabu.science/paper/10.1038%2Fs41467-026-75680-8

## Key claims

1. 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.
   - Support: Tentative
   - Type: Primary finding
   - Basis: comparator-dependent benchmarks do not establish across-workflow superiority in speed and memory
   - Location: Abstract
2. Dreamlet processes pseudobulk for 1000 donors across eight cell types for 4.1 million cells in 31 min using only 26 Gb of memory
   - Support: Supported
   - Type: Secondary
   - Basis: single-thread benchmark uses resampled donors and reports averages without timing dispersion
   - Location: Results, Computational performance on large-scale datasets; Fig. 2A, B
3. Of the methods that control the false positive rate across a range of sample sizes, Dreamlet is the only one that can model random effects
   - Support: Supported
   - Type: Secondary
   - Basis: batch-preserving label permutations support the tested-method comparison within one brain dataset
   - Location: Results, Statistical performance on large-scale datasets; Fig. 2E
4. [In Alzheimer's disease microglia,] PTPRG stands out with a log2 fold change of 1.59 and p-value of 2.94e-28
   - Support: Supported
   - Type: Secondary
   - Basis: postmortem observational association lacks orthogonal validation and may partly reflect microglial state composition
   - Location: Results, Large effect up-regulation of PTPRG in microglia of Alzheimer’s disease cases; Fig. 6A
5. microglia have 1037 differentially expressed genes passing the study-wide 5% FDR cutoff
   - Support: Supported
   - Type: Secondary
   - Basis: observational case–control analysis supports a cohort-specific discovery count rather than a causal disease effect
   - Location: Results, Large effect up-regulation of PTPRG in microglia of Alzheimer’s disease cases; Fig. 6A

## Strengths

- **Scalable mixed models address donor dependence** Dreamlet combines donor-level pseudobulk analysis with precision-weighted mixed models for repeated measurements and multiplexed batches. This extends established workflows to complex cohort designs rather than treating cells as independent biological replicates.
  - Dimension: contribution
  - Location: Results, Dreamlet workflow; Fig. 1; Methods, Precision-weighted linear mixed models
- **Complementary validation anchors the computational workflow** Validation combines an independent simulation pipeline, batch-preserving disease-label permutations and established computational comparators. The new 299-donor cohort adds technical replicates and genotype-based checks of donor identity.
  - Dimension: methodological rigour
  - Location: Fig. 2D, E; Fig. 5; Methods, Alignment and demultiplexing
- **Power limitations inform the biological interpretation** The Discussion explains how unequal nuclei counts and measurement precision alter differential-expression power across cell clusters. It also identifies within-sample distributions and continuous gradients that pseudobulk aggregation cannot preserve.
  - Dimension: positioning
  - Location: Fig. 5G, H; Discussion, differential-power and pseudobulk-limitations paragraphs

## Limitations

- **Broad superiority wording exceeds benchmark distinctions** The conclusion implies broader speed, memory and false-positive advantages than the Results establish. Other pseudobulk methods also control false positives, DE fitting speed is competitive with edgeR and DESeq2, and aggregation advantages vary by comparator and sample size.
  - Dimension: reporting
  - Location: Abstract; Fig. 2A–E; Discussion, concluding paragraph
- **Scaling evidence relies on resampled donors** The 1,000-donor aggregation benchmark samples donors with replacement from the underlying cohort. Ten single-thread runs provide average timings and memory usage, but no dispersion is reported.
  - Dimension: methodological rigour
  - Location: Fig. 2A, B; Methods, Performance comparison for computing pseudobulk
- **Calibration evidence covers limited validation settings** The real-data false-positive benchmark uses one brain dataset, while the reported ground-truth simulations use simple designs. These checks support the tested settings but leave greater uncertainty about calibration and power across other complex designs.
  - Dimension: methodological rigour
  - Location: Results, Statistical performance on large-scale datasets; Fig. 2D, E

## Reliability checks

Caveats (4 of 4 checks).

- **Results check:** Clean. The reported numbers and statistics are consistent with the stated results.
- **Reporting and transparency:** Noted. Ethics approval, competing interests, and data and code repositories are declared. A cross-reference in the prostate cancer application names TB rather than the preceding COVID-19 application, without changing the reported results.
  - Conflict of interest: declared · Ethics approval: declared · Data availability: declared · Code availability: declared
- **Reference integrity:** Noted
  - 74 references in manuscript
  - 67 of 74 checkable references found in an index
  - Could not be verified automatically — queued for manual review:
    - Zeng, H. What is a cell type and how to define it? Cell 185, 2739–2755 (2022). (Crossref bibliographic search unavailable (search could not be performed) — no conclusion drawn about this reference)
    - Tanay, A. & Regev, A. Scaling single-cell genomics from phenomenology to mechanism. Nature 541, 331–338 (2017). (Crossref bibliographic search unavailable (search could not be performed) — no conclusion drawn about this reference)
    - Finucane, H. K. et al. Partitioning heritability by functional annotation using genome-wide association summary statistics. Nat. Genet. 47, 1228–1235 (2015). (Crossref bibliographic search unavailable (search could not be performed) — no conclusion drawn about this reference)
    - Oelen, R. et al. Single-cell RNA-sequencing of peripheral blood mononuclear cells reveals widespread, context-specific gene expression regulation upon pathogenic exposure. Nat. Commun. 13, 3267 (2022). (Crossref bibliographic search unavailable (search could not be performed) — no conclusion drawn about this reference)
    - Kang, H. M. et al. Multiplexed droplet single-cell RNA-sequencing using natural genetic variation. Nat. Biotechnol. 36, 89–94 (2018). (Crossref bibliographic search unavailable (search could not be performed) — no conclusion drawn about this reference)
    - Huang, Y., McCarthy, D. J. & Stegle, O. Vireo: Bayesian demultiplexing of pooled single-cell RNA-seq data without genotype reference. Genome Biol. 20, 273 (2019). (Crossref bibliographic search unavailable (search could not be performed) — no conclusion drawn about this reference)
    - COMBAT Consortium. A blood atlas of COVID-19 defines hallmarks of disease severity and specificity. Cell 185, e58 (2022). (Crossref bibliographic top candidate ratio 0.50 — no plausible match)
  - References not verified automatically are resolved manually within 24 hours.

## Post-publication record

- **Status:** Clean
- **Summary:** No retraction or expression of concern found in Crossref (includes Retraction Watch).
- **Last checked:** 2026-11-22
