AI-Powered Imaging Enables Faster Lung Disease Treatment

Current treatments can only slow disease progression and do not reverse the fibrotic damage. The unpredictable nature of IPF and lack of effective therapeutic strategies make it a difficult disease to study and treat. Now, a newly developed deep learning algorithm offers a powerful approach to interpreting disease data and identifying potential treatments, marking a significant step forward in understanding IPF and other complex diseases.

The tool, known as UNAGI (unified in-silico cellular dynamics and drug screening framework), was developed by researchers at Yale School of Medicine (New Haven, CT, USA) in collaboration with other institutions. It is a deep generative neural network designed to uncover disease-specific biological patterns and propose drug candidates with minimal human oversight. UNAGI works by analyzing vast datasets to model disease progression at a cellular level. Using sequencing data from 230,000 cells, the AI system builds virtual representations of cells and their disease states.

What sets UNAGI apart from other models is its disease-informed design—it not only identifies genes and regulatory networks involved in disease progression but also integrates this information back into the model for continuous refinement. This embedded iterative learning allows the tool to autonomously interpret new data and test different drugs, eliminating the need for manual re-training. The system also pulls from a database of thousands of drugs with known mechanisms of action to predict which compounds might be effective for a specific disease.

UNAGI was initially trained using IPF data collected from lung tissue samples obtained during transplant surgeries. The tissue was sliced and analyzed to represent different stages of disease, enabling the Yale team to catalogue gene expression in single cells and create a pulmonary fibrosis single-cell atlas. The model identified relevant regulatory networks and classified disease stages, then screened thousands of drugs to identify eight with potential anti-fibrotic effects. One of these was nifedipine, a calcium channel blocker commonly used for hypertension, which UNAGI flagged as a potential treatment for IPF. Laboratory validation using IPF-modeled lung tissue slices confirmed UNAGI’s prediction—nifedipine successfully inhibited scar tissue formation.

This research, published in Nature Biomedical Engineering, underscores the transformative potential of merging AI with single-cell sequencing to unlock new avenues for treating IPF and other progressive diseases. UNAGI’s ability to analyze complex datasets, model disease trajectories, and autonomously identify therapeutic candidates could reshape clinical research by accelerating the discovery of effective treatments. Although initially applied to IPF, the tool has also shown promise in analyzing other disease states, including aging and COVID-19, suggesting its wide applicability across biomedical research and drug development.

Related Links:
Yale School of Medicine