GraLTR-LDA

GraLTR-LDA predicts potential associations between long non-coding RNAs (lncRNAs) and diseases to prioritize candidate lncRNA-disease pairs for downstream experimental validation.

Key Features:

• Multi-source graph construction: Builds homogeneous and heterogeneous graphs from integrated multi-source biological data to represent lncRNAs, diseases, and related entities.
• Graph auto-encoder with attention: Applies a graph auto-encoder combined with an attention mechanism to extract embedded features that capture patterns and dependencies in the graphs.
• Learning to Rank with feature crossing: Integrates extracted embeddings into a Learning to Rank framework using feature crossing statistical strategies to prioritize diseases associated with specific lncRNAs.

Scientific Applications:

• Genomics research: Predicts lncRNA-disease associations to guide discovery of novel therapeutic targets and diagnostic markers.
• Personalized medicine: Prioritizes candidate lncRNA-disease pairs to support development of individualized diagnostics and treatments.
• Experimental prioritization: Ranks potential associations to focus experimental validation on the most promising lncRNA-disease links.

Methodology:

Graph construction: building homogeneous and heterogeneous graphs from multi-source biological data. Feature extraction: using a graph auto-encoder with an attention mechanism to derive embedded features. Ranking prediction: integrating these features into a Learning to Rank framework with feature crossing strategies to predict and rank disease associations for lncRNAs.