GSLRDA

GSLRDA predicts associations between noncoding RNAs (ncRNAs) and drug resistance to identify potential mechanisms underlying therapeutic resistance.

Key Features:

• Graph-Based Modeling: Models known ncRNA–drug resistance associations as a bipartite graph with nodes representing ncRNAs and drugs.
• lightGCN: Uses a light Graph Convolutional Network (lightGCN) to learn vector representations of ncRNAs and drugs from the bipartite graph topology.
• Self-Supervised Learning: Generates multiple views for each ncRNA and drug node via data augmentation and leverages self-supervised learning to improve representations without labels.
• Contrastive Learning: Applies contrastive learning between augmented views to refine node representations and enhance discrimination of associations.
• Prediction Mechanism: Predicts ncRNA–drug resistance associations by computing the inner product of learned vector representations.
• Performance: Achieves an Area Under the Curve (AUC) of 0.9101, outperforming eight other state-of-the-art models.

Scientific Applications:

• ncRNA–Drug Resistance Prediction: Predicts potential associations between ncRNAs and drug resistance to reveal candidate resistance mechanisms.
• Drug Development Support: Identifies ncRNA targets that may inform strategies to overcome therapeutic resistance during drug development.
• Case Study Analysis: Applies predicted associations to case studies involving specific drugs to assess practical relevance.

Methodology:

Construct a bipartite graph from known ncRNA–drug associations; apply lightGCN to learn vector representations; use data augmentation to create multiple views and self-supervised learning with contrastive learning to refine representations; predict associations via the inner product of learned vectors.