Computational Learning of microRNA-Based Prediction of Pouchitis Outcome After Restorative Proctocolectomy in Patients With Ulcerative Colitis. Issue 10 (20th February 2021)

• Record Type: Journal Article
• Authors: Morilla, Ian; Uzzan, Mathieu; Cazals-Hatem, Dominique; Colnot, Nathalie; Panis, Yves; Nancey, Stéphane; Boschetti, Gilles; Amiot, Aurélien; Tréton, Xavier; Ogier-Denis, Eric; Daniel, Fanny
• Is Part Of: Inflammatory bowel diseases. Volume 27:Issue 10(2021)
• Page Start: 1653
• Page End: 1660
• Publication Date: 2021-02-20
• Subjects: prognostic biomarker -- pouchitis -- deep learning; Inflammatory bowel diseases -- Periodicals; Colitis, Ulcerative -- Periodicals; Crohn Disease -- Periodicals; Inflammatory Bowel Diseases -- Periodicals; 616.344
• DOI: 10.1093/ibd/izab030
• Languages: English
• ISSNs: 1078-0998

Abstract

Abstract: Background: Ileal pouch-anal anastomosis (IPAA) is the standard of care after total proctocolectomy for ulcerative colitis (UC). However, inflammation often develops in the pouch, leading to acute or recurrent/chronic pouchitis (R/CP). MicroRNAs (miRNA) are used as accurate diagnostic and predictive biomarkers in many human diseases, including inflammatory bowel diseases. Therefore, we aimed to identify an miRNA-based biomarker to predict the occurrence of R/CP in patients with UC after colectomy and IPAA. Methods: We conducted a retrospective study in 3 tertiary centers in France. We included patients with UC who had undergone IPAA with or without subsequent R/CP. Paraffin-embedded biopsies collected from the terminal ileum during the proctocolectomy procedure were used for microarray analysis of miRNA expression profiles. Deep neural network–based classifiers were used to identify biomarkers predicting R/CP using miRNA expression and relevant biological and clinical factors in a discovery cohort of 29 patients. The classification algorithm was tested in an independent validation cohort of 28 patients. Results: A combination of 11 miRNA expression profiles and 3 biological/clinical factors predicted the outcome of R/CP with 88% accuracy (area under the curve = 0.94) in the discovery cohort. The performance of the classification algorithm was confirmed in the validation cohort with 88% accuracy (area under the curve = 0.90). Apoptosis, cytoskeletal regulation by Rho GTPase, and fibroblast growth factor signaling were the most dysregulated targets of the 11 selected miRNAs. Conclusions: We developed and validated a computational miRNA-based algorithm for accurately predicting R/CP in patients with UC after IPAA.