Early disease correlated protein detection using early response index (ERI)

Abstract

Finding disease correlated proteins with significant expression changes before the clinical onset stage of disease is of great interest to researchers for developing early disease diagnosis biomarkers. Since disease correlated proteins have relatively low level of abundance change at early stages, it is hard to find them using existing bioinformatic tools in high throughput data, which has limited dynamic range and significant noise. Most existing biomarker discovery algorithms can only detect proteins with high abundance changes, and early disease diagnostic markers are frequently missed. We proposed a new ranking score called early response index (ERI) for prioritizing disease correlated protein as potential early diagnostic markers. Rather than classification accuracy, ERI estimates the average classification accuracy improvement achievable by proteins when they are combined with other proteins as features of classifiers. ERI is more sensitive to abundance changes than other ranking statistics. In a validation study to detect proteins with sustained expression changes from the pre-clinical onset to the clinical onset stage of multiples sclerosis using a mouse model, the proposed algorithm outperforms other tested algorithms in both sensitivity and specificity.