Fully connected autoencoder and convolutional neural network with attention-based method for inferring...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Abnormal long noncoding RNA (lncRNA) expression is linked to human diseases, making disease-associated lncRNA identification crucial for understanding pathogenesis.
• Current prediction methods often overlook the intricate topology within meta-paths involving lncRNAs, diseases, and microRNAs (miRNAs).

Purpose of the Study:

• To propose a novel method, ACLDA, for inferring potential disease-related lncRNA candidates.
• To enhance the integration of topological information from heterogeneous networks composed of lncRNA, disease, and miRNA nodes.

Main Methods:

• Constructed a heterogeneous graph integrating lncRNA, disease, and miRNA similarities, associations, and interactions.
• Employed fully connected autoencoders for low-dimensional feature extraction and attention mechanisms for enhanced feature and meta-path learning.
• Utilized convolutional neural networks to encode local topologies from multiple meta-path perspectives.

Main Results:

• ACLDA demonstrated superior performance compared to existing state-of-the-art prediction methods.
• Experimental results confirmed ACLDA's effectiveness in identifying potential disease-related lncRNAs.
• Case studies on breast, lung, and colon cancers validated ACLDA's predictive capabilities.

Conclusions:

• ACLDA offers a powerful approach for discovering disease-associated lncRNAs by effectively integrating complex network topology and multi-level features.
• The method advances the understanding of disease pathogenesis by identifying novel lncRNA-disease associations.
• ACLDA shows significant potential for application in clinical diagnostics and therapeutic target identification.