A machine learning approach for predicting familial and sporadic disease cases based on clinical symptoms: introduction of a new dataset

Objective: Neurofibromatosis type 1 (NF1) is a common yet complex neurogenetic disorder characterized by a highly variable clinical presentation, influenced by both genetic and environmental factors. While its genetic basis is well understood, the variability in symptoms among patients presents significant challenges for diagnosis and management. This study focuses on examining the differences in clinical features between sporadic and familial NF1 cases. Additionally, it evaluates the potential of machine learning techniques to predict sporadic NF1 cases based on clinical symptoms, offering insights into how computational approaches can complement traditional diagnostic methods. Methods: A retrospective analysis was conducted on the medical records of 241 NF1 patients, including 121 sporadic and 120 familial cases. The frequency of various clinical features, such as Lisch nodules, pseudoarthrosis, and hypertension, was compared between the groups. analysis of variance (ANOVA) was used to identify the most important features distinguishing sporadic cases from familial ones. Furthermore, multiple machine learning algorithms, including k-nearest neighbors, artificial neural networks, support vector machines, decision trees, and XGBoost, were employed to predict sporadic cases based on the identified features. Results: Among the machine learning models tested, the XGBoost algorithm demonstrated the highest predictive accuracy at 62.86%, indicating moderate reliability in identifying sporadic cases. Despite this limitation, the analysis revealed significant differences in clinical manifestations between the two groups. These differences suggest that shared genetic modifiers may play a critical role in shaping the observed genotype-phenotype relationship in NF1. Conclusion: This study represents the first detailed comparison of a broad spectrum of clinical symptoms between sporadic and familial NF1 cases. While machine learning models showed only moderate success in prediction, the findings provide valuable insights into the phenotypic variability of NF1 and underscore the importance of larger, more diverse datasets for improving predictive accuracy. These results hold significant potential for guiding personalized diagnostic and therapeutic strategies for NF1 patients.