Interpretable artificial intelligence model for accurate identification of medical conditions using immune repertoire.

This article focuses on the development and evaluation of VDJMiner, an interpretable artificial intelligence (AI) model designed to identify underlying medical conditions and predict prognosis in COVID-19 patients based on their T-cell receptor (TCR) V(D)J gene segment usage. Using immune repertoire data from over 1,400 COVID-19 patients, VDJMiner accurately classified eight medical conditions—including cancer, autoimmune disease, diabetes, asthma, chronic kidney disease, congestive heart failure, coronary artery disease, and chronic obstructive pulmonary disease—with an average area under the receiver operating characteristic curve (AUC) of 0.961, and predicted severe COVID-19 with an AUC of 0.922. The model also demonstrated an 85.7% accuracy in forecasting patient response to tocilizumab treatment. Compared to other machine learning methods such as TabNet and support vector machines, VDJMiner showed superior and more stable performance, while providing interpretable insights into disease-associated V(D)J gene segments, consistent with prior studies. The source code and web server for VDJMiner are publicly accessible, enabling further research and application in personalized diagnosis and prognosis of COVID-19 and potentially other diseases.