LAUSR

INTRODUCTION Wilson disease (WD) is characterized by a wide variety of clinical manifestations. Our study aimed to correlate genotype with clinical and radiological features in Indian WD patients. METHODS We conducted a descriptive observational study in a tertiary care neurology referral center of eastern India over a period of 2 years. Demographic data collection, clinical examination and relevant investigations were done for all WD patients meeting the inclusion criteria. Based on previous reports of mutation hotspots for WD in Eastern India, we performed PCR-Sanger sequencing of selected exons of ATP7B gene. To understand the role of each of these covariates on the occurrence of common mutation, we applied a logistic regression as well as random forest in a supervised learning framework. RESULTS Fifty-two WD patients were included in the study. c.813C > A (p.C271X) was the commonest identified mutation. The statistical methods applied to our data-set reveal the most important features for predicting common mutation or its absence. We also found that the state-of-the-art classification algorithms are good at predicting the absence of common mutation (with true positive rates being 0.7647 and 0.8823 for logistic classifier and random forest, respectively), but predicting the occurrence remains a harder modeling challenge. CONCLUSIONS WD patients in eastern India have significant genotypic and phenotypic diversity. Statistical methods for binary classification show some early promise of detecting common mutations and suggest important covariates, but further studies with larger samples and screening of remaining exons are warranted for understanding the full genetic landscape of Wilson disease.