DAPNEML: Disease-diet associations prediction in a NEtwork using a machine learning based approach

Abstract

Generic notions about associations between certain diseases and diets are quite popular, but there are many evidences of other unknown disease-diet associations in literature that need to be fully explored. Such associations are currently being studied by medical researchers through meta-analysis or other prospective studies limiting it to a certain population or area. This study aims to use a combined view of such associations from literature for predicting unknown associations using advanced computational techniques including Network Analysis and Machine Learning. Disease-Diet Associations Prediction in a NEtwork using Machine Learning (DAPNEML) is an approach designed to curate known disease-diet and diet-diet associations data from literature, visualize and integrate the data in the form of a network, extract features from these complex interdependencies using network algorithms and predict unknown associations using machine learning. The predictions are performed in two phases, with the first predicting if an association exists between disease-diet whereas the second predicting the nature of its association (diet is harmful or helpful for a disease). Accuracies achieved in phase 1 and phase 2 are 83% and 76% respectively. The proposed approach can be of great help for researchers and biomedical professionals in constructing diet based disease progressions.