Idiopathic pulmonary fibrosis (IPF) is one of the harder pulmonary indications. Median survival from diagnosis sits in the three-to-five-year range, and the two approved drugs — nintedanib and pirfenidone — slow disease progression but do not stop it. The biology cuts across activated fibroblasts laying down matrix, damaged alveolar epithelium failing to repair, and a chronic immune environment that supports both. Each of those compartments is a different cell-type story.
This case study walks through what the Euretos AI Platform reveals when those compartments are queried together.
A “Related Cells” search for idiopathic pulmonary fibrosis in the Euretos AI Platform returns 1,065 cell-type associations across the indexed PubMed corpus. The top of that ranking, by reference count, is consistent with the histological hallmarks of the disease:
| Rank | Cell type | References |
|---|---|---|
| 1 | Fibroblasts | 2,520 |
| 2 | Epithelial cells | 1,503 |
| 3 | Myofibroblasts | 1,104 |
| 4 | Macrophages | 1,051 |
| 5 | Lymphocytes | 660 |
| 6 | Neutrophils | 550 |
| 7 | Stem cells | 509 |
| 8 | Leukocytes | 480 |
| 9 | Squamous epithelial cells | 454 |
| 10 | Alveolar macrophages | 442 |
| 11 | T-lymphocytes | 371 |
| 12 | Pneumocytes | 322 |
| 13 | Monocytes | 304 |
| 14 | Mesenchymal stem cells | 262 |
| 15 | Lung fibroblasts | 259 |
Two of the top three are mesenchymal — fibroblasts and the activated myofibroblast subset that produces most of the disease-driving matrix. The next tier is immune. The epithelial signal sits between them, reflecting the alveolar damage that initiates and sustains the fibrotic response. A few entries deeper down become biologically more specific: alveolar macrophages at rank 10 (442 references), and type-II pneumocytes — the alveolar progenitor cells whose dysfunction is now considered an intrinsic driver of IPF pathogenesis — at rank 21 with 234 references.
This ranking is the literature-derived view: it tells the user which cell types the field has spent time studying in IPF. The next analytical step is to resolve each compartment into the specific cell-type subsets the recent single-cell atlases have characterised.
The classic histological hallmark of IPF is the fibroblastic focus — a region of activated fibroblasts and myofibroblasts depositing collagen at the boundary between damaged alveolar epithelium and the underlying interstitium. Single-cell RNA sequencing of fibrotic lung has identified a CTHRC1-expressing fibroblast subset that emerges in fibrotic lungs and expresses the highest levels of collagen of any lung-resident cell (Tsukui et al. 2020). These cells are concentrated within the fibroblastic foci themselves, and their disease-relevance has been confirmed in immunostaining and adoptive-transfer experiments. They are the cellular engine of scar formation.
A target-discovery query restricted to the CTHRC1-fibroblast subset emphasises matrix-deposition machinery, integrin-mediated activation, Wnt-pathway components, and the specific transcription factors associated with the disease-emergent fibroblast state. Targets here are about disabling the matrix engine.
A landmark single-cell IPF atlas of more than 312,000 cells from 32 IPF patients and matched controls identified an aberrant basaloid cell population that sits at the edge of fibroblastic foci and co-expresses basal epithelial, mesenchymal, senescence, and developmental markers (Adams et al. 2020). These cells are not present in healthy lung. They appear to be a stalled-repair state — alveolar epithelium that has begun to transdifferentiate but never resolves. Reviews of subsequent single-cell IPF work describe the same population as a convergent feature across atlases (Justet, Zhao & Kaminski 2022).
The platform’s cell-type ranking also surfaces type-II pneumocytes (the canonical alveolar progenitor cells, also known as AT2 cells), epithelial cells of the alveolus, and respiratory epithelial cells as distinct entries. Each of these is a refinement of the broader “epithelial cells” entry at rank 2: they let a target-discovery query sit at the cell-type subset, rather than the tissue-compartment, level. The aberrant basaloid population sits inside the alveolar-epithelial subset and reflects the disease-emergent state within it.
A target-discovery query restricted to the aberrant basaloid subset shifts toward genes implicated in epithelial-to-mesenchymal transition, senescence pathways, and developmental signalling. Targets here are about preventing or resolving the stalled epithelial state, rather than blocking the matrix-producing cells directly.
Profibrotic monocyte-derived macrophage populations are expanded in IPF lung and have been confirmed across multiple datasets (Justet, Zhao & Kaminski 2022). Their gene programmes overlap with macrophage states seen in tissue repair more broadly, but in IPF the overlap shifts in a way that favours sustained matrix deposition rather than resolution.
The platform’s ranking refines this further: alveolar macrophages (the lung-resident, tissue-imprinted population) sit at rank 10 with 442 references, monocytes at rank 13 with 304 references, and the more recent literature on alternatively activated (M2-like) macrophages, T-helper subsets, and dendritic cells appears further down the list — all available as cell-type filters when the target-discovery query is being constructed.
A query restricted to the profibrotic macrophage subset surfaces tissue-repair-skewed myeloid factors and the receptor-ligand pairs by which these cells communicate with adjacent fibroblasts and epithelium. Targets here are about breaking the supporting immune environment.
Inside the AI Platform, the IPF target-discovery view begins at the disease level — the integrated knowledge graph anchors on the IPF MeSH and ontology terms — and then resolves down through tissue (lung), tissue compartment (parenchyma), and cell type. The platform’s cell-type filter, mapped to the same cell ontology used by the published atlases, lets a user move directly from “IPF” to “aberrant basaloid cells in IPF lung” or “CTHRC1-expressing fibroblasts in IPF lung.”
The candidate-target ranking changes meaningfully at each step. At the bulk-lung level, ranking is dominated by canonical fibrosis biology — TGF-β pathway components, integrin α-v subunits, collagen genes themselves. Most of these have been pursued in clinic and most have failed or returned modest effects. At each cell-type-resolved level, the ranking shifts toward biology specific to that compartment, with the supporting evidence stack — animal model evidence, GWAS support, expression context, paralog landscape, druggability — surfaced for each candidate without leaving the Search interface.
This is the same target-discovery query, asked four ways, with four different shortlists.
The Euretos AI Platform does not rank IPF candidates from scratch. It integrates published genetic associations, cross-disease MeSH-mapped evidence, perturbation data, and literature-extracted relationships, and then applies the cell-type filter on top. For a researcher investigating IPF, that means the day-to-day analytical question is not “which cell-type atlas should I download and re-process” but “which cell-type compartment am I asking about right now.” The integration work is already done.
The AI Platform does not pretend to provide the clinical answer. Translational success in IPF will depend on which compartment is the right point of intervention for which patient subgroup. What the platform provides is a defensible, traceable, cell-type-resolved view of the candidate space — built on the same integrated knowledge graph that has powered Euretos’s target-discovery work since the platform launched.
The next case study in this series will look at psoriasis, where the cell-type story is different again.