Multifactor Dimensionality Reduction for the Analysis of Obesity in a Nutrigenetics Context

摘要

The current work aims to study within a nutrigenetics context the multifactorial trait beneath obesity. To this end, the use of parallel Multifactor Dimensionality Reduction (pMDR) is investigated towards the identification of ⅰ) factors that have an impact to obesity onset solely or interacting with each other and ⅱ) rules that describe the interactions among them. Data have been obtained from a large scale nutrigenetics study and each subject, characterized as normal or overweight based on Body Mass Index (BMI), is featured a 63-dimensional vector describing his/her genetic variations and nutritional habits. pMDR method was used to reduce the initial set of factors into subsets that can classify a subject into either normal or overweight with a certain accuracy and are further used by corresponding prediction models. Results showed that pMDR selected factors associated to obesity and constructed predictive models showing a good generalization ability. Rules describing interactions of the selected factors were extracted, thus enlightening the classification mechanism of the constructed model.