A protease activity-based machine-learning approach as a complementary tool for conventional diagnosis of diarrhea-predominant irritable bowel syndrome

Abstract

Irritable bowel syndrome (IBS) has no clinically accepted biomarkers even though it affects a large number of individuals worldwide. To address this lack of understanding, we evaluated peptidase activity in fecal samples from 35 patients with diarrheal IBS without symptom exacerbation (IBS-n) and 35 healthy subjects using a library of 384 fluorescent enzymatic substrate probes. IBS-n patients had high trypsin-like peptidase activity for cleavage of C-terminal lysine and arginine residues and low elastase-like activity for cleavage of C-terminal serine and glycine residues. These fluorescent probe library data, together with diagnostic machine-learning techniques, were able to accurately predict IBS-n. This approach can be used to diagnose diseases where no clinically accepted biomarkers exist, in which fecal enzyme activity is altered and also suggests that the development of new therapies targeting enzyme activities is possible.

IntegrIBS: Evaluating the Potential and Challenges of Building a Robust IBS Classifier with Integrated Microbiome Data

“…Fukui et al [7] demonstrated the utility of machine learning-based gut microbiome analysis in identifying IBS patients. Tanaka et al [8] also used a machine-learning approach to diagnose IBS subtypes, showing the utility of advanced algorithms. However, these studies often relied on single datasets, limiting their generalizability.…”