No abstract

No abstract

Background Many miRNA-based diagnostic models have been constructed to distinguish diseased individuals. However, due to the inherent differences across different platforms or within multi-center data, the models usually fail in the generalization for medical application. Results Here, we proposed to use the within-sample expression ratios of related miRNA pairs as markers, by utilizing the internal miRNA: miRNA interactions. The ratio of the expression values between each miRNA pair turned out to be more stable cross multiple data source. Moreover, we adopted the genetic algorithm to solve the curse of dimensions when exploring the features. Conclusions The application results on three example datasets demonstrated that the expression ratio of interacting miRNA pair is a promising type of biomarker, which is insensitive to batch effects and has better performance in disease classifications.