TERRA: A Tissue World Model Pretrained on 112M Cells from Spatial Transcriptomics

mo_lotfollahi · x · 2026-08-06

TERRA is a tissue world model pretrained on 112M cells from spatial transcriptomics, spanning 20 tissues and 26 disease conditions, with 2/3 newly generated. Using a JEPA-like approach, it learns by prediction in embedding space to capture multi-scale tissue logic, aiming for a universal representation of human tissue.

Related event: TERRA: Tissue World Model Trained on 112 Million Cells(4 posts)→

Original post →

More from Research

Research channel →