Research in Progress | Rising Star Seminar

The tumor immune microenvironment (TIME) is critical for cancer progression and immunotherapy response, but spatial proteomics methods to study TIME remain expensive and difficult to scale. In this talk, I will present GigaTIME, a multimodal AI framework that translates routine H&E pathology slides into virtual spatial proteomics maps. Trained on 40 million paired H&E–mIF cells across 21 proteins, GigaTIME was applied to over 14,000 patients across 24 cancer types to generate a population-scale virtual TIME atlas from 50+ hospitals and 1000+ clinics of the Providence health system. This enabled large-scale discovery of spatial protein activation patterns associated with biomarkers, staging, survival, and patient stratification - demonstrating how multimodal AI can unlock scalable medical discovery from real-world clinical data.

Speaker Bio: Jeya Maria Jose Valanarasu, PhD, is a Senior Researcher at Microsoft Research in the Real-World Evidence group, where he develops multimodal AI systems for healthcare and biomedical discovery using population-scale real-world data. His research focuses on applications in radiology, pathology, and oncology, with an emphasis on virtual patient/population modeling, spatial medicine, and AI-driven clinical discovery. Previously, he was a postdoctoral researcher at Stanford University working with Andrew Ng and Curt Langlotz on medical foundation models and synthetic data generation. He earned his PhD and MS from Johns Hopkins University under the supervision of Vishal M. Patel. His work has been recognized with honors including the MICCAI Young Scientist Impact Award and the Amazon Research Fellowship, and has appeared at leading venues such as Cell, Nature, CVPR, ICLR, ECCV, and MICCAI.