The article introduces ROSIE, a novel deep-learning framework designed to computationally generate multiplex immunofluorescence (mIF) staining data from standard, inexpensive hematoxylin and eosin (H&E) histopathology images. H&E staining is common but lacks molecular specificity, which mIF provides, albeit at a high cost and with complex procedures. ROSIE is trained on a massive dataset of over 1,300 paired H&E and mIF samples to impute the expression and localization of dozens of proteins. Validation results show that the predicted biomarkers are highly effective for detailed cell phenotyping, including distinguishing hard-to-identify immune cells like B and T lymphocytes, and are useful for identifying tissue structures. The authors argue that this method has significant potential to enhance clinical workflows by providing rich molecular information without requiring expensive mIF assays.
References:
