Pathway analysis using xgboost classification in biomedical data

Abstract

Given the fact that our biological existence is rooted in a complex system within our cells with thousands of interactions among genes and metabolites, research community in biological and medical fields have shifted their interest to network-based approaches. This complexity is imprinted in networks which encode the relationships among system’s components. This evolution has also led to the generation a new research fields, the Network Medicine, a combination of Network Science and Systems Biology applied to human diseases. Meanwhile, cutting-edge approaches towards this direction are the subpathway-based methods, identifying “active subpathways” - in the form of local sub-structures within pathways - related with a case under study. Based on this, we propose a classification scheme based on XGBoost, a recent tree-based classification algorithm, in order to detect the most discriminative pathways related with a disease. Subsequently, we extract subpathways and rank them with regard to their ability to correctly classify samples from different experimental conditions. Our method is demonstrated on an aging gene expression dataset providing evidences that XGBoost outperforms other well-known classification methods in biological data, while results provided by our method include several established as well as recently reported longevity-associated pathways. © 2018 Association for Computing Machinery.