The abstract is read at the publisher; the summary is JClub's.
Graph-guided Flow Matching (GFM) predicts post-perturbation single-cell transcriptomes for unseen single-gene Perturb-seq perturbations, outperforming baselines with up to 30x faster training.
- Why it matters: Exhaustively mapping the full perturbation space with Perturb-seq is impractical, and existing models struggle to generalize to unseen perturbations.
- What they did: GFM combines biological knowledge graphs with latent flow matching in a generative framework, evaluated across multiple large-scale Perturb-seq datasets for single-gene and two-gene perturbation prediction.
- The result: GFM improves prediction of unseen perturbation responses, remains robust with limited training data, and in chronic myeloid leukemia links HSP90AA1 inactivation to transcriptional consequences and target candidates not seen during training.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.