Dynamic Graph Prior-Guided Latent Adaptation for Generalizable Single-Cell Perturbation Response Prediction.

DGP-LA enhances generalization in single-cell perturbation response prediction, achieving up to 79.1% improvement in Pearson Delta over existing models for unseen cell types.

• Why it matters: Accurately predicting transcriptional responses in unobserved cellular contexts is crucial for advancing single-cell perturbation modeling, yet current models struggle with generalization and biological interpretability.
• What they did: The authors developed DGP-LA, a dynamic graph-prior latent adaptation framework that refines biological priors through perturbation-conditioned sparse graph refinement and integrates them as gated latent corrections within a cell-state-conditioned transition model, tested on Tahoe and NeurIPS datasets with key performance metrics.
• The result: DGP-LA significantly improves response prediction accuracy, gene recovery, and pathway analysis, enabling more generalizable and biologically interpretable models, with source code available online for reproducibility.