Noise-Consistent Hypergraph Autoencoder Based on Contrastive Learning for Cancer ceRNA Association Prediction in...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Competitive endogenous RNA (ceRNA) networks are crucial for understanding noncoding RNA roles in complex diseases.
• Traditional graph models struggle with long-range dependencies and noise in biological networks.
• Existing hypergraph models have limitations in handling graph-level and node-level noise.

Purpose of the Study:

• To propose a Noise-Consistent hypeRgraph AutoEncoder (NCRAE) framework for robust node embeddings in ceRNA networks.
• To enable precise prediction of cancer-related ceRNA biomarkers.
• To improve predictive performance in the presence of noise.

Main Methods:

• NCRAE utilizes a multiview contrastive learning strategy with graph-level and node-level corruption.
• A noise consistency loss constraint is incorporated to mitigate contrastive learning biases and enhance noise resistance.
• Hypergraph convolution and Fourier KAN techniques are employed for effective node embedding learning.

Main Results:

• NCRAE demonstrates superior performance compared to existing methods, particularly under noisy conditions.
• The framework achieves robust node embedding learning in ceRNA regulatory networks.
• Experimental results validate NCRAE's robustness and predictive capability for cancer biomarker discovery.

Conclusions:

• NCRAE offers a powerful tool for identifying cancer-related ceRNA biomarkers.
• The proposed method effectively addresses noise challenges in ceRNA network analysis.
• NCRAE shows significant practical value in cancer biomarker prediction and discovery.