Graph Convolutional Network and Contrastive Learning Small Nucleolar RNA (snoRNA) Disease Associations (GCLSDA): Predicting snoRNA–Disease Associations via Graph Convolutional Network and Contrastive Learning

作者：Liangliang Zhang, Ming Chen, Xiaowen Hu, Lei Deng · 发表于：International Journal of Molecular Sciences · 年份：2023 · DOI：10.3390/ijms241914429 · 被引用次数：15 · 研究领域：Cancer-related molecular mechanisms research、RNA modifications and cancer、RNA Research and Splicing

Small nucleolar RNAs (snoRNAs) constitute a prevalent class of noncoding RNAs localized within the nucleoli of eukaryotic cells. Their involvement in diverse diseases underscores the significance of forecasting associations between snoRNAs and diseases. However, conventional experimental techniques for such predictions suffer limitations in scalability, protracted timelines, and suboptimal success rates. Consequently, efficient computational methodologies are imperative to realize the accurate predictions of snoRNA–disease associations. Herein, we introduce GCLSDA—graph Convolutional Network and contrastive learning predict snoRNA disease associations. GCLSDA is an innovative framework that combines graph convolution networks and self-supervised learning for snoRNA–disease association prediction. Leveraging the repository of MNDR v4.0 and ncRPheno databases, we construct a robust snoRNA–disease association dataset, which serves as the foundation to create bipartite graphs. The computational prowess of the light graph convolutional network (LightGCN) is harnessed to acquire nuanced embedded representations of both snoRNAs and diseases. With careful consideration, GCLSDA intelligently incorporates contrast learning to address the challenging issues of sparsity and over-smoothing inside correlation matrices. This combination not only ensures the precision of predictions but also amplifies the model’s robustness. Moreover, we introduce the augmentation technique of random noise t...