A Hybrid Feature Selection Method for Classifying Metagenomic Data in Relation to Inflammatory Bowel Disease

Abstract

Due to the recent advances in high throughput metagenomic sequencing technologies, Microbial abundance profiles of environmental samples have become publicly available. Increasing number of metagenomic studies has associated the imbalance of bacterial abundance to health and disase state of the host. This suggests utilizing the bacterial profiles as a diagnostic tool to identify the bacterial-related disease state of individuals. However, the high dimensional nature of metagenomic datasets renders this process a challenging task. Therefore, an efficient framework that enables accurate classification of metagenomic samples belonging to different classes is of central important. In this work, a hybrid feature selection technique that combines the advantages of filter and wrapper feature selection algorithms is proposed. The experimental results demonstrate that the proposed algorithm outperforms widely used feature selection techniques in terms of classification accuracy and provide a significant reduction in the computation time.