GL4SDA: Predicting snoRNA-disease associations using GNNs and LLM embeddings

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Small nucleolar RNAs (snoRNAs) are crucial for cellular processes, with growing evidence linking them to various diseases.
• Accurate identification of snoRNA-disease relationships is vital for understanding their biological roles and therapeutic potential.

Purpose of the Study:

• To develop a novel computational approach, GL4SDA, for predicting associations between small nucleolar RNAs (snoRNAs) and diseases.
• To leverage Graph Neural Networks (GNNs) and Large Language Models (LLMs) for improved prediction accuracy.

Main Methods:

• GL4SDA utilizes heterogeneous graph structures to model complex biological interactions between snoRNAs and diseases.
• The method incorporates snoRNA secondary structures and disease embeddings from LLMs to create rich node features.
• A GNN model with high-performing layers is designed to maximize predictive results based on these integrated features.

Main Results:

• GL4SDA demonstrated superior performance in link prediction tasks compared to existing state-of-the-art graph-based predictors.
• The model effectively integrates structural snoRNA features with semantic disease embeddings, enhancing predictive power.
• Explainable AI methods identified key snoRNA features, validating findings through cancer disease case studies.

Conclusions:

• GL4SDA offers a powerful new tool for exploring snoRNA-disease associations.
• The approach highlights the potential of combining structural and semantic data for biological network inference.
• Findings underscore the practical applicability of advanced computational methods in biomedical research.