Overview
- Developed with Providence and the University of Washington, the Cell paper describes an AI that generates virtual multiplex immunofluorescence from standard pathology images.
- Training on paired H&E and mIF data across 21 protein channels produced about 40 million matched cells for supervision.
- GigaTIME processed 14,256 Providence cases spanning 24 cancer types and 306 subtypes, creating roughly 299,376 virtual images and identifying more than 1,200 protein–clinical associations.
- External evaluation on 10,200 TCGA tumors found strong concordance across datasets, with performance surpassing CycleGAN on 15 of 21 protein channels.
- The tool is publicly available on Microsoft Foundry Labs and Hugging Face for research use, with authors noting limitations in cohort diversity and variable translation quality and emphasizing that clinical deployment needs further validation.