SVMBN: A new method for predicting metabolite-disease associations based on biological networks and machine learning

Abstract

Identifying metabolite-disease associations is of paramount significance. With the advancement of research, computational methods have surpassed traditional experiments in efficiency. Nevertheless, current computational methods often overlook the integration of multiomics data, and the performance of the predictive models used is limited. To address these limitations, we propose the SVMBN algorithm for predicting metabolite-disease associations. The proposed approach involves the following steps: First, six similarity calculation methods are employed to construct the metabolite similarity network and the disease similarity network separately. Second, the metabolite and disease similarity networks are combined to obtain the original link features. Third, nonnegative Matrix Factorization (NMF) is applied to extract effective features from the original features, thereby reducing noise. Finally, Support Vector Machine (SVM) is utilized to predict potential associations between metabolites and diseases. Experimental results demonstrate that the SVMBN algorithm achieves an average AUC of 0.98 in 5-fold cross-validation, indicating its superiority over other methods. Furthermore, case studies prove that the SVMBN algorithm can accurately forecast the relationships between metabolites and diseases.