An Interpretable Classification Model Using Gluten-Specific TCR Sequences Shows Diagnostic Potential in Coeliac...

Area of Science:

• Immunology
• Gastroenterology
• Computational Biology

Background:

• Coeliac disease (CeD) is a significant global health issue, often underdiagnosed due to limitations in current diagnostic methods.
• The gold standard for CeD diagnosis involves invasive endoscopic biopsies and gluten challenges, which are patient-unfriendly and subject to diagnostic variability.
• There is a critical need for a more objective, less invasive, and accurate diagnostic test for CeD.

Approach:

• This study reviewed existing gluten-specific T-cell receptor (TCR) sequences.
• An interpretable machine learning model was developed and trained using TCR repertoire data from CD4+ T cells of CeD patients and controls.
• The model's diagnostic potential was validated on an independent dataset, including patients on a gluten-free diet (GFD).

Key Points:

• The developed machine learning model achieved 100% accuracy in training and 80% accuracy in testing for CeD diagnosis.
• The model demonstrated diagnostic capability even in patients adhering to a gluten-free diet.
• Twenty specific CD4+ TCR sequences with high diagnostic potential for CeD were identified.

Conclusions:

• Identifying gluten-specific TCRs offers a promising avenue for a less subjective and more accessible CeD diagnostic test.
• This novel approach could potentially eliminate the need for gluten challenges in CeD diagnosis.
• The identified TCR sequences may form the basis of a new objective diagnostic tool for coeliac disease.