Attentional multi-level representation encoding based on convolutional and variance autoencoders for lncRNA-disease...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Abnormalities in long non-coding RNAs (lncRNAs) are linked to various human diseases, making disease-related lncRNA identification critical for understanding disease pathogenesis.
• Current prediction methods often overlook crucial node attributes and pairwise feature distributions in lncRNA-disease relationships, limiting their predictive power.

Purpose of the Study:

• To propose VADLP, a novel prediction model designed to extract, encode, and adaptively integrate multi-level representations for enhanced lncRNA-disease association prediction.
• To address limitations in existing methods by incorporating node attributes, pairwise topology, and feature distribution for a more comprehensive analysis.

Main Methods:

• Constructed a triple-layer heterogeneous graph integrating similarities and correlations among lncRNAs, diseases, and microRNAs (miRNAs).
• Defined and extracted three key representations: node attributes (via embedding), pairwise topology (via random walk and convolutional autoencoder), and feature distribution (via variance autoencoder).
• Developed an attentional representation-level integration module to adaptively fuse these representations for final prediction.

Main Results:

• The VADLP model significantly outperformed six state-of-the-art lncRNA-disease prediction models on a public dataset.
• Ablation studies confirmed the significant contributions of the three integrated representations to the model's performance.
• Demonstrated improved recall rates, indicating high efficacy in identifying true disease-related lncRNAs within top-ranked candidates.

Conclusions:

• VADLP provides a powerful and statistically significant advancement in predicting lncRNA-disease associations by effectively integrating multi-level data representations.
• The model's ability to discover potential disease-related lncRNAs was further validated through case studies on three specific cancers.