Direct discovery · Biology · Medicine

Expert-guided AutoDiscovery surfaces a lobular breast-cancer immune signal validated in tissue

An immune-checkpoint signal in invasive lobular breast cancer, supported by an independent cohort and multiplex immunofluorescence.

Summary

An unguided 100-hypothesis run yielded no clinically useful finding. After expert feedback, a 500-hypothesis run highlighted immune-checkpoint expression in lobular cancer. Researchers checked the signal in METABRIC and examined 12 lobular tumors and six normal-tissue samples.

AI role

Generated and tested hypotheses in TCGA breast-cancer data after an oncologist annotated results from an unproductive unguided run.

Narrative role

Links AI-generated hypotheses to additional biological evidence while showing that expert steering and laboratory follow-up remained essential.

Caveat

Preprint with small tissue samples and observational comparisons. Immune markers do not establish response to immunotherapy; the study does not measure a causal productivity gain. Event date is the preprint posting, after the August 27 announcement.