Artificial Intelligence Algorithm Based on Genetics to Predict the Response to Interferon Beta Treatment in Multiple Sclerosis Patients

Multiple sclerosis (MS) is a chronic inflammatory disease of the central nervous system (CNS) affecting nearly 3 million people worldwide. Although its etiology and pathogenesis are not clearly established, it is likely the result of complex interactions between genetic and environmental factors. The MS autoimmune etiology has been the goal of therapeutic approaches for patients. IFN-β is one of the most prescribed disease-modifying therapies in MS patients. However, IFN-β is partially effective, and a significant proportion of patients partially respond or do not respond to this treatment. Machine learning (ML) is a subset of artificial intelligence (AI) focusing on developing data-driven computational models to enhance specific tasks. In recent years, unsupervised learning methods, such as hierarchical clustering and K-Means, have been applied in the MS study for classification tasks. Although these methods are relatively easy to implement, they rarely provide the best solution due to the large number of arbitrary decisions, and they perform only when the dataset contains clusters of similar sizes and no notable outliers. Therefore, in this paper, we propose an AI algorithm including a fuzzy expert system, based on the opinion of a neurology expert, to improve the efficiency in the assignment of unknown class labels associated to the response to IFN-β in MS patients, and a genetic algorithm (GA) to optimize the hyperparameter tuning of a deep learning (DL) model trained with genetic biomarkers to estimate whether the patients are potential candidates to be treated with this therapy. The experimental results show the fuzzy system achieved a 80% of classification efficiency against 64% of a conventional hierarchical clustering method. Furthermore, the artificial neural network (ANN) model, whose hyperparameters were optimized by the GA, achieved an 1.0 accuracy against 0.8 of a multi-layer perceptron (MLP), whose hyperparameters were adjusted by a conventional tuning method.