DELPHAI predicts heterogeneous perturbation responses with learned single-cell fitness

DELPHAI predicts heterogeneous cellular responses to perturbation with outperforming all baseline methods by leveraging learned cell fitness filtering and gene-space retrieval.

• Why it matters: Understanding diverse cell responses is crucial for scalable in silico screening and mechanistic insights, but current models are limited by mass conservation issues and information loss during projection.
• What they did: The approach involves a novel model, DELPHAI, which filters cells based on learned fitness and retrieves gene-space information to overcome latent space bottlenecks, tested across two benchmark frameworks.
• The result: DELPHAI consistently outperforms existing methods, provides explainability through cell-type-specific survival predictions, and operates without requiring prior biological knowledge.