The study presents the model UNAGI (“Unsupervised Neural Analysis of Gene Interactions”)—a generative deep-learning method that, for the first time, uses longitudinal single-cell transcriptomic data to reconstruct cellular dynamics over the course of chronic diseases. While traditional analyses provide static snapshots, UNAGI enables a time-resolved view of disease progression. From high-dimensional transcriptomic data, the model generates so-called disease-informed embeddings that map functional cellular states according to disease stages.

Using data from idiopathic pulmonary fibrosis (IPF), the researchers demonstrated that UNAGI can accurately distinguish between early, intermediate, and late disease phases at the cellular level. The model identified fibroblast subpopulations associated with progressive tissue remodeling and traced their transcriptional programs throughout the disease. These insights allow the identification of regulatory networks that are crucial for the transition from reversible to irreversible fibrotic states.

A particularly innovative aspect is the integration of cellular-dynamics modeling with an in silico drug-screening approach. Based on the generated disease profiles, UNAGI can predict which pharmacological substances might modulate specific pathological cell states. Potential antifibrotic candidates were experimentally tested in human precision-cut lung slices, where their predicted effects were validated through measurable reductions in fibrotic markers.

“With approaches like these, we bring cellular mechanisms—and ultimately therapeutic options—much closer to the patient,” emphasizes BREATH PI Jonas Schupp. The findings highlight the potential of modern AI-supported methods not only to describe disease mechanisms at the single-cell level but also to make them therapeutically actionable.

The publication emerged from close international collaboration and underscores the importance of globally networked research for advancing pneumology. For BREATH and the German Center for Lung Research (DZL), it marks an important step toward data-driven, precise, and patient-oriented lung research.

The original publication can be found here.

Text: BREATH/AB

Foto: privat