RNA fingerprinting, a statistical framework that learns denoised perturbation signatures from single-cell reference dictionaries and probabilistically maps query cells to causal perturbations with uncertainty quantification, enables accurate perturbation assignment at genome scale, identifies context-specific regulators including RPL10 and RPL24 as novel mediators of p53 activation under ribosomal stress, reveals dose-dependent off-target effects in putatively selective drugs, and uncovers heterogeneous cytokine-driven immune responses in vivo that are undetectable by conventional analysis methods.

Single-cell perturbation dictionaries provide systematic measurements of how cells respond to genetic and chemical perturbations, and create the opportunity to assign causal interpretations to observational data. Here, we introduce RNA fingerprinting, a statistical framework that maps transcriptional responses from new experiments onto reference perturbation dictionaries. RNA fingerprinting learns denoised perturbation “fingerprints” from single-cell data, then probabilisti-cally assigns query cells to one or more candidate perturbations while accounting for uncertainty. We benchmark our method across ground-truth datasets, demonstrating accurate assignments at single-cell resolution, scalability to genome-wide screens, and the ability to resolve combinatorial perturbations. We demonstrate its broad utility across diverse biological settings: identifying context-specific regulators of p53 under ribosomal stress, characterizing drug mechanisms of action and dose-dependent off-target effects, and uncovering cytokine-driven B cell heterogeneity during secondary influenza infection in vivo. Together, these results establish RNA fingerprinting as a versatile framework for interpreting single-cell datasets by linking cellular states to the underlying perturbations which generated them.