A decision support system based on multi-sources information to predict piRNA–disease associations using stacked autoencoder.

Despite experimental efforts to map out the human interactome, the continued high data complexity, heterogeneity, and combined explosion limit our ability to understand the molecular roots of human disease. Computational tools provide a promising alternative when identifying biologically significant, yet unmapped associations between PIWI-interacting RNA (piRNA) and diseases. However, few predictors for large-scale prediction of potential disease-associated piRNAs have been proposed to provide effective candidates for future biomedical research now. In this study, we proposed a new computational model based on multi-source information and stacked autoencoder, called MSRDA, to predict potential piRNA–disease associations on a large scale. In particular, we introduce stacked autoencoders into the field of piRNA–disease association prediction for the first time. In the fivefold cross-validation, the MSRDA achieves the average area under the curve of 0.9184 ± 0.0015. To further evaluate the performance of MSRDA, we compared models without feature denoising and models with different types of feature representations. Moreover, the proposed method is optimal compared to related works. In conclusion, MSRDA is an effective tool that provides new impetus for uncovering the root causes of human disease. [ABSTRACT FROM AUTHOR]