What do single-cell models already know about perturbations?

What do single-cell models already know about perturbations?
Andreas Bjerregaard, Iñigo Prada-Luengo, Vivek Das and Anders Krogh
Genes 16 (12) 2025; also ICLR MLGenX 2025
https://www.mdpi.com/2073-4425/16/12/1439

Abstract: Background: Virtual cells are embedded in widely used single-cell generative models. Nonetheless, the models' implicit knowledge of perturbations remains unclear. Methods: We train variational autoencoders on three gene expression datasets spanning genetic, chemical, and temporal perturbations, and infer perturbations by differentiating decoder outputs with respect to latent variables. This yields vector fields of infinitesimal change in gene expression. Furthermore, we probe a publicly released scVI decoder trained on the CELL × GENE Discover Census ($\sim!5.7$M mouse cells) and score genes by the alignment between local gradients and an empirical healthy-to-disease axis, followed by a novel large language model-based evaluation of pathways. Results: Gradient flows recover known transitions in Irf8 knockout microglia, cardiotoxin-treated muscle, and worm embryogenesis. In the pretrained Census model, gradients help identify pathways with stronger statistical support and higher type 2 diabetes relevance than an average expression baseline. Conclusions: Trained single-cell decoders already contain rich perturbation-relevant information that can be accessed by automatic differentiation, enabling in-silico perturbation simulations and principled ranking of genes along observed disease or treatment axes without bespoke architectures or perturbation labels.

Running the project

To reproduce the results, follow these steps.

1. Download the following datasets:

• C. elegans embryogenesis [1], GEO: GSE126954. Files to be placed in data/celegans/ with: cell annotations as cells.csv, transcript names as features.csv and counts as counts.mtx.
• Irf8-cKO mouse brains [2], GEO: GSE128855 ("Full aggregate"). Files to be placed in data/mousebrain/ with: cell barcodes as cells.tsv, transcript names as features.tsv, counts as counts.mtx and metadata as annotations.csv.
• Cardiotoxin-induced injury in mice [3], ArrayExpress: E-MTAB-9715. Files to be placed in data/cardiotoxin/ with: cell barcodes as cells.mtx_cols, transcript names as features.mtx_rows , counts as counts.mtx and design file as annotations.tsv.

2. For each dataset, run the relevant preprocessing script in the preprocessing folder.

3. Finally, the .ipynb notebooks can be run to generate the results.

Citation

To cite the work or codebase:

@article{bjerregaard2025single,
  title={What do single-cell models already know about perturbations?},
  author={Bjerregaard, Andreas and Prada-Luengo, I{\~n}igo and Das, Vivek and Krogh, Anders},
  journal={Genes},
  volume={16},
  number={12},
  pages={1439},
  year={2025},
  publisher={Multidisciplinary Digital Publishing Institute}
}

References

1. Packer, Jonathan S., et al. "A lineage-resolved molecular atlas of C. elegans embryogenesis at single-cell resolution." Science 365.6459 (2019): eaax1971.
2. Van Hove, Hannah, et al. "A single-cell atlas of mouse brain macrophages reveals unique transcriptional identities shaped by ontogeny and tissue environment." Nature neuroscience 22.6 (2019): 1021-1035.
3. Takada, Naoki, et al. "Galectin-3 promotes the adipogenic differentiation of PDGFRα+ cells and ectopic fat formation in regenerating muscle." Development 149.3 (2022): dev199443.