Related Experiment Video
Updated: Jan 10, 2026
06:58
Collection and Extraction of Saliva DNA for Next Generation Sequencing
Published on: August 27, 2014
39.9K
Interpretable machine learning applied to high-dimensional salivary proteomics accurately classifies pediatric
Brittany T Rupp1, Joaquin Reyna2, Ally Giunta3,4
1Department of Oral and Craniofacial Molecular Biology, Virginia Commonwealth University, Richmond VA., USA.
Medrxiv : the Preprint Server for Health Sciences
|November 24, 2025
Summary
This study introduces a novel salivary proteomic and AI approach for non-invasive pediatric inflammatory bowel disease (IBD) classification. It identifies a minimal protein signature for accurate diagnosis, improving early therapeutic decisions.
Area of Science:
- Gastroenterology
- Proteomics
- Artificial Intelligence
Background:
- Inflammatory bowel diseases (IBD), including Crohn's disease (CD), ulcerative colitis (UC), and IBD-unclassified (IBD-U), are chronic gastrointestinal inflammatory conditions.
- Current diagnostic and monitoring methods for pediatric IBD are invasive, costly, and time-consuming, hindering timely management.
- There is a need for non-invasive, accurate, and rapid methods for IBD classification and monitoring.
Purpose of the Study:
- To apply high-dimensional salivary proteomics integrated with interpretable artificial intelligence/machine learning (AI/ML) for pediatric IBD classification.
- To identify a minimal protein signature in saliva for differentiating pediatric IBD subtypes (CD, UC, IBD-U).
- To inform therapeutic decision-making through accurate and timely IBD diagnosis.
Main Methods:
- Analysis of unstimulated saliva from pediatric patients with CD, UC, and IBD-U using the NULISAseq Inflammation Panel (250 proteins).
- Identification of a minimal discriminative protein signature using logistic regression and recursive feature elimination.
- Validation of the model's performance on independent follow-up samples and assessment of patient-specific protein contributions using SHapley Additive exPlanations (SHAP).
Main Results:
- A 14-protein signature was identified, including chemokines, cytokines, receptors, ligands, and structural proteins.
- The AI/ML model achieved 96.2% accuracy in classifying first-visit samples and 86.4% in follow-up testing.
- SHAP analysis revealed patient-specific drivers and indicated that IBD-U cases align biologically with either CD or UC profiles.
Conclusions:
- The integration of salivary proteomics and interpretable AI/ML enables accurate, non-invasive classification of pediatric IBD using minimal biomarker sets.
- This approach provides a scalable framework for future longitudinal monitoring of IBD.
- The findings support earlier and more precise therapeutic interventions for pediatric IBD patients.
