A tissue time machine to forward-predict single-timepoint data applied to tuberculosis granulomas

Recent advancements have enhanced our understanding of infections; however, many detailed datasets only capture a single point in time post-infection, hindering the generation of longitudinal insights necessary to predict infection outcomes. For instance, when Mycobacterium tuberculosis (Mtb) initiates a pulmonary infection, a complex system of immune and bacterial interactions unfolds, culminating in granuloma formation.

Yet, these single snapshots in time require further elaboration on a long-term scale. To address this challenge, researchers have developed a novel approach called the tissue time machine. This method is applied to predict outcomes of non-human primate granulomas from initiation to endpoint using single timepoint tissue datasets, which include high-dimensional spatial features.

The tissue time machine involves identifying digital partners of granulomas, simulating models most closely resembling real granulomas from a virtual library comprising over 50,000 such models. By analyzing these digital partners, the researchers were able to predict infection outcomes for 22 out of 28 granulomas. Furthermore, the study revealed that Mtb-specific, IFN-gamma-producing T cell recruitment rates significantly influence the uncertainty associated with the remaining 6 granulomas.

This innovative method holds the potential to be applied to any system that allows for mechanistic predictions over time, even when dealing with heterogeneous, incomplete, and high-dimensional datasets.