Cross-modal transfer enables zero-shot prediction of drug responses across cell types

GRAFT enables zero-shot prediction of drug-induced transcriptional responses across cell types, improving response correlation and reconstruction versus competing methods, especially for context-dependent drugs.

• Why it matters: Drug-response prediction often lacks drug measurements in the target cellular context, and transferring information across perturbation modalities and contexts is a key gap.
• What they did: The framework combines genetic-to-chemical transfer with cross-context modeling, benchmarking on LINCS and external contexts, and uses GRAFT-Adapt with a small number of target-context drug measurements.
• The result: GRAFT-Atlas spans 13 cellular contexts and 7,856 drugs, producing 102,128 predictions and capturing pharmacological neighborhoods and context-dependent changes; across ten iPSC lines, it identifies candidate inter-donor differences supported by measured genetic perturbations.

AI summary of the abstract · may contain errors · not medical advice · read the paper.