Research
CellPert: Graph-based variational learning with biological priors for cross-dataset perturbation prediction
Patterns · By Qing Wang, Sen Guo, Jialu Liang, Weizhou Zhang, Jie Xu & Qianqian Song ·
The Brief
- CellPert is a graph-based AI framework that incorporates known gene interactions to predict cellular responses.
- It transfers perturbation effects learned from LINCS L1000 to unseen single cells in the Tahoe100M dataset.
- The authors report improved predictive accuracy for cross-dataset perturbation prediction.
Why it matters
Predicting how cells respond to perturbations across different datasets could reduce the need for costly experiments, and using known gene interactions as biological priors may make such predictions more interpretable.
Who’s affected
- computational biologists
- researchers working on single-cell perturbation studies
- #perturbation prediction
- #single-cell genomics
- #graph learning
- #gene interactions
Wang et al. introduce CellPert, a graph-based AI framework that incorporates known gene interactions to predict cellular responses across datasets. By transferring perturbation effects from LINCS L1000 to unseen single cells in Tahoe100M, CellPert improves predictive accuracy…