VGAELDA

VGAELDA predicts associations between long non-coding RNAs (lncRNAs) and human diseases using variational inference and graph autoencoder architectures to identify candidate lncRNA–disease links relevant to prognosis, diagnosis, and therapy.

Key Features:

• End-to-End Model: Integrates variational inference with graph autoencoders to predict unknown lncRNA-disease associations.
• Variational Graph Autoencoders (VGAE): Infers low-dimensional embeddings from high-dimensional features of lncRNAs and diseases.
• Graph Autoencoders for Label Propagation: Propagates labels through known associations to enhance prediction of new links.
• Variational Expectation Maximization Algorithm: Alternately trains autoencoders to optimize learning and improve robustness of predictions.
• Co-Training Framework: Co-trains multiple autoencoders to address a geometric matrix completion problem and capture complementary representations.
• Low-Dimensional Representation Learning: Reduces high-dimensional biological data to compact representations for downstream association inference.

Scientific Applications:

• lncRNA–disease association prediction: Identifies and prioritizes potential links between lncRNAs and human diseases.
• Candidate prioritization for prognosis, diagnosis and therapy: Ranks novel lncRNA–disease associations to inform potential clinical research targets.
• Representation learning for biological features: Produces low-dimensional embeddings of lncRNAs and diseases to support downstream analyses.

Methodology:

Learning low-dimensional representations from high-dimensional lncRNA and disease features; alternating training of variational graph autoencoders and graph autoencoders using a variational expectation maximization algorithm within a co-training framework; label propagation through known associations and application of variational inference techniques.