gGATLDA: lncRNA-disease association prediction based on graph-level graph attention network

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Long non-coding RNAs (lncRNAs) regulate gene expression and are implicated in human diseases.
• Identifying lncRNA-disease associations (LDAs) is crucial for disease diagnosis, treatment, and prognosis.
• Experimental identification of LDAs is costly, time-consuming, and inefficient, necessitating computational approaches.

Purpose of the Study:

• To develop an efficient and accurate computational method for predicting LDAs.
• To leverage graph neural networks for enhanced LDA prediction.
• To provide a valuable tool for researchers investigating lncRNA functions in disease.

Main Methods:

• Proposed a novel computational method, gGATLDA, based on a graph-level graph attention network.
• Extracted enclosing subgraphs for each lncRNA-disease pair.
• Constructed feature vectors integrating lncRNA and disease similarity as node attributes.
• Trained a graph neural network (GNN) model to predict potential LDA scores.

Main Results:

• gGATLDA achieved superior performance in AUC, AUPR, accuracy, and F1-Score compared to state-of-the-art methods via five-fold cross-validation.
• Case studies demonstrated the method's effectiveness in identifying lncRNAs associated with breast, gastric, prostate, and renal cancers.
• The model successfully predicted potential lncRNA-disease associations.

Conclusions:

• The gGATLDA method is a highly effective computational approach for predicting potential LDAs.
• The findings support the utility of graph attention networks in bioinformatics for disease-related RNA research.
• This method offers a significant advancement in the computational identification of LDAs.