CGSDA: Inferring snoRNA-disease associations via ChebNetII and GatedGCN

Yongfu Zou¹Yusong Lu^1*Shanghui Lu¹Zhanliang Wei²Le Li³Shuilin Liao⁴Ting Zeng⁴Yi Zhang⁵Miao Rui³

• ¹Hechi University, Hechi, China
• ²The Second Nanning People's Hospital, Nanning, China
• ³Zunyi Medical University - Zhuhai Campus, Zhuhai, China
• ⁴Macau University of Science and Technology, Macau, China
• ⁵Nanning Normal University, Nanning, China

Recent biomedical studies have highlighted the pivotal role of non-coding RNAs (ncRNAs) in gene regulatory networks, where they influence gene expression, cellular function, and the onset and progression of various diseases. Among these, small nucleolar RNAs (snoRNAs), a prominent class of small ncRNAs, have attracted considerable research attention over the past two decades. Initially recognized for their involvement in rRNA processing and modification, snoRNAs are now understood to contribute to broader biological processes, including the regulation of disease mechanisms, maintenance of cellular homeostasis, and development of targeted therapeutic strategies. With ongoing advancements, snoRNAs are increasingly regarded as promising candidates for novel therapeutic agents in cancer, neurodegenerative disorders, endocrine conditions, and cardiovascular diseases. Consequently, there is a growing demand for efficient, cost-effective, and environment-independent approaches to study snoRNAs, which has driven the adoption of computational methodologies in this domain. In this work, we propose a novel predictive framework, CGSDA, which integrates a ChebNetII convolutional network with a gated graph sequence neural network to identify potential snoRNA–disease associations. The model begins by constructing a snoRNA–disease association network, embedding residual mechanisms into both modules to effectively capture the representations of snoRNAs and diseases. These representations are then fused and dimensionally reduced, after which the refined embeddings are fed into a predictor to generate association predictions. Experimental evaluation demonstrates that CGSDA consistently outperforms baseline models in predictive accuracy. Ablation experiments were conducted to assess the contribution of each module, confirming that all components substantially enhance overall performance and validating the robustness of the proposed method. Furthermore, case studies on lung cancer and breast cancer showed that 10 out of the top 15 and 12 out of the top 15 predicted snoRNA-disease associations Zou et al. SDA prediction by CGSDA were validated by existing literature, respectively, confirming the model's effectiveness in identifying potential novel snoRNA-disease associations. The implementation of CGSDA, along with relevant datasets, is publicly available at: https://github.com/cuntjx/CGSDA.