Identification of Important Biological Pathways for Ischemic Stroke Prediction through a Mathematical Programming Optimisation Model-DIGS

Stroke ranks second after heart disease as a cause of disability in high-income countries and as a cause of death worldwide. Identifying the biomarkers of ischemic stroke is possible to help diagnose stroke cases from non-stroke cases, as well as advancing the understanding of the underlying theory of the disease. In this study, a mathematical programming optimisation framework called DIGS is applied to build a phenotype classification and significant pathway inference model using stroke gene expression profile data. DIGS model is specifically designed for pathway activity inference towards supervised multi-class disease classification and is proved has great performance among the mainstream pathway activity inference methods. The highest accuracy of the prediction on determining stroke or non-stroke samples reaches 84.4% in this work, which is much better than the prediction accuracy produced by currently found stroke gene biomarkers. Also, stroke-related significant pathways are inferred from the outputs of DIGS model in this work. Taken together, the combination of DIGS model and expression profiles of stroke has better performance on the discriminate power of sample phenotypes and is capable of effective in-depth analysis on the identification of biomarkers.