In a major advance for understanding and treating idiopathic pulmonary fibrosis (IPF), researchers at Yale School of Medicine and international collaborators have developed an AI system that learns how lung scarring progresses, and proposes drug candidates to stop it.
Published in Nature Biomedical Engineering, the tool, called UNAGI (unified in-silico cellular dynamics and drug screening framework), can model disease-specific regulatory networks and suggest therapies by analyzing single-cell sequencing data.
“The model looks for regulation—what characterizes and regulates changes—and then, using known drug databases, also suggests treatments,” said Naftali Kaminski, MD, senior author and professor of pulmonary medicine at Yale.
IPF: an incurable fibrotic disease
IPF is a progressive and fatal lung condition where the lung’s peripheral tissue becomes fibrotic, impairing breathing. Despite its severity, the cause remains unknown. Current drugs, pirfenidone and nintedanib, only slow disease progression and can come with side effects. They do not reverse the underlying tissue scarring.
Unlike previous AI tools, UNAGI is disease-informed. It models the disease by first learning how gene expression and regulatory pathways shift between healthy and fibrotic cells. Then, using drug databases, it tests thousands of compounds virtually, seeking those that can halt or reverse these pathological processes.
AI uncovers hidden drivers of fibrosis
The Yale team began with gene expression data from over 230,000 cells taken from fibrotic lungs at different stages of disease. With collaborators at KU Leuven and McGill University, they fed the data into UNAGI.
The model rapidly identified key regulators of fibrosis, mapped how different lung cells evolve during disease progression, and highlighted eight candidate drugs, one of which is already approved for IPF.
Intriguingly, the AI also flagged a calcium channel blocker, nifedipine, typically used for hypertension, as having anti-fibrotic potential. The researchers tested this in lab-grown lung tissue and confirmed that it blocked scar formation, matching the AI’s prediction.
“UNAGI is hitting pathways we did not think about before,” Kaminski said.
A platform for other diseases
Though designed for IPF, UNAGI was also tested on COVID-19 data and aging-related tissue data, suggesting it could be repurposed for a variety of complex conditions. Unlike conventional models, it can autonomously refine itself when exposed to new data, cutting down on time and cost.
“The model evolves to understand more and more about the disease,” said co-author Jun Ding of McGill University.
The convergence of AI and single-cell sequencing is reshaping disease modeling. In the case of IPF, UNAGI not only accelerates drug discovery but opens a new window into the molecular drivers of fibrosis.
Although nifedipine may not become a frontline therapy, the ability to uncover previously overlooked therapeutic pathways marks a leap forward in pulmonary research.
“We have a convergence of both very sophisticated AI-based analytical methods and the ability to generate data that is high-resolution enough to actually make these observations,” Kaminski said.
