A Stacked Autoencoder-Based miRNA Regulatory Module Detection Framework

• Main Authors: Yi Yang, Yan Song
• Format: Article
• Language: English
• Published: Atlantis Press 2019-01-01
• Series: International Journal of Computational Intelligence Systems
• Subjects: Module detection; Intimacy; K-means; Stacked autoencoder
• Online Access: https://www.atlantis-press.com/article/125914770/view

MicroRNA regulatory module (MRM) plays an important role in the study of microRNA synergism. To detect MRMs, researchers have developed a number of related methods in the preceding decades. However, some existing methods are stochastic or specific to a certain situation. In this paper, we presented a novel deep ensemble framework called DeMosa to identify MRM for different cancers. In the proposed framework, we integrated stacked autoencoders and K-means method to detect MRMs in high-dimensional complex biological networks. We tested our method on synthetic data and three types of cancer data sets. In the synthetic data, we found DeMosa is superior to existing three methods SNMNMF, Mirsynergy, and bi-cliques merging (BCM) on clustering accuracy, stability, and module quality, while in the cancer datasets, DeMosa is more adaptable in different situations than the counterparts. In addition, we applied Kaplan–Meier survival analysis to predict several MRMs as potential prognostic biomarkers in cancers.