Global Shapley Explanations and LIME on HLA-B27 Positivity in Ankylosing Spondylitis Patients

Abstract

Abstract Ankylosing spondylitis (AS), an autoimmune disease, has the HLA-B27 gene in more than 90% of its patients. This study investigated the ability of health parameters to predict the presence of the HLA-B-27 gene and clinical and demographic data used in diagnosing AS. For this purpose, various classification models were evaluated, and the best-performing RFC model was selected. In addition, the model's predictions are understood and explained using XAI techniques such as SHAP and LIME. The model development results show that the RFC model performs best (Accuracy:0.75, F1 Score:0.74, Recall:0.75, Precision:0.75, Brier Score:0.25, AUC: 0.76), and XAI techniques provide the ability to explain the decisions of this model. Among the health parameters, WBC, Hematocrit, uric acid, and gender were found to show the strongest association with HLA-B-27. This study aims to understand the genetic predisposition of AS and to illuminate the potential of XAI techniques in medical diagnosis. The study's strengths include comprehensive model evaluation, explainability of model decisions, and revealing the relationship between health parameters and HLA-B-27. In addition, this study considered ethical dimensions like the confidentiality of personal health data and the privacy of patients.