gGATLDA

gGATLDA predicts associations between long non-coding RNAs (lncRNAs) and diseases by applying a graph-level graph attention network to subgraph representations.

Key Features:

• Graph-Level Graph Attention Network: Employs a graph-level graph attention mechanism within a graph neural network (GNN) to model lncRNA-disease relationships.
• Subgraph Extraction: Extracts enclosing subgraphs for each lncRNA-disease pair to capture local network structure and relevant interactions.
• Feature Vector Construction: Constructs feature vectors by integrating lncRNA similarity and disease similarity as node attributes within the extracted subgraphs.
• GNN Training: Trains the GNN model on the extracted subgraphs and corresponding feature vectors to learn patterns indicative of associations.
• Performance Evaluation: Reports predictive performance using area under the receiver operating characteristic curve (AUC), area under the precision recall curve (AUPR), accuracy, and F1-Score under five-fold cross-validation.

Scientific Applications:

• lncRNA-disease association prediction: Predicts potential associations between lncRNAs and diseases to prioritize candidate lncRNAs for further study.
• Cancer association identification: Identifies lncRNAs associated with cancers including breast cancer, gastric cancer, prostate cancer, and renal cancer.
• Mechanism and biomarker research: Supports investigation of molecular underpinnings, diagnostic biomarkers, and targets for therapeutic development.

Methodology:

Extract enclosing subgraphs for each lncRNA-disease pair; construct node feature vectors by integrating lncRNA similarity and disease similarity; train a graph-level graph attention network (GNN) on these subgraphs and feature vectors; evaluate predictions with five-fold cross-validation reporting AUC, AUPR, accuracy, and F1-Score.