| Abstrakt: |
Background/Aims Fibrosis in Ulcerative Colitis (UC) leads to strictures often requiring surgery. Current biomarkers and imaging fail to predict immune-mediated fibrotic progression. This study develops a machine learning model integrating single-cell multi-omics and spatial transcriptomics to predict fibrosis in UC. Identifying early fibrotic signatures aims to improve precision medicine in inflammatory bowel disease (IBD), enabling proactive intervention and reducing surgeries. Methods Single-cell RNA sequencing (scRNA-seq) and spatial transcriptomics data from the Gene Expression Omnibus (GEO) and the Human Cell Atlas (HCA) were utilized, including samples from 500 Ulcerative Colitis patients and 200 controls. Epigenetic data, including ATAC-seq (chromatin accessibility) and H3K27ac histone modifications, were sourced from the Encyclopedia of DNA Elements (ENCODE) and the International Human Epigenome Consortium (IHEC). Immune phenotyping and cytokine signaling data were integrated from mass cytometry studies available through the Immunology Database and Analysis Portal (ImmPort). A deep learning model combining graph convolutional networks (GCNs) for spatial analysis and recurrent neural networks (RNNs) for temporal dynamics was developed to predict fibrosis onset within two years. Performance was measured using area under the receiver operating characteristic curve (AUROC), sensitivity, specificity, and precision-recall metrics, with cross-validation ensuring robustness. Results The model achieved an AUROC of 0.91 (95% CI: 0.87-0.94), sensitivity of 85.4% (95% CI: 82.1-88.3), and precision of 88.7% (95% CI: 85.3-91.2). Elevated expression of COL1A1, TGF-β, and ACTA2 in stromal cells (p<0.001) and spatial clustering of profibrotic macrophages were associated with a 22% increased risk of fibrosis (p<0.001). Histone modification H3K27ac in T cells was linked to fibrotic progression (OR: 2.5, 95% CI: 1.9-3.2, p<0.001). Incorporating spatial transcriptomics improved model accuracy by 15% (p<0.0001). Conclusion This machine learning model integrating single-cell multi-omics and spatial transcriptomics provides accurate fibrosis prediction in UC, enabling early, non-invasive risk assessment for improved management. [ABSTRACT FROM AUTHOR] |
