Node-adaptive graph Transformer with structural encoding for accurate and robust lncRNA-disease association...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Long noncoding RNAs (lncRNAs) play crucial roles in gene regulation, cell differentiation, and cancer development.
• Predicting lncRNA-disease associations deepens understanding of disease pathogenesis and supports therapeutic development.

Purpose of the Study:

• To introduce an innovative computational model, NAGTLDA, for predicting unknown lncRNA-disease associations.
• To leverage graph neural networks and attention mechanisms for improved association prediction.

Main Methods:

• Utilized node-adaptive feature smoothing (NAFS) for local feature learning.
• Employed Structural Deep Network Embedding (SDNE) for encoding network structure.
• Integrated Transformer modules with multi-headed attention for capturing global associations and network structure coding for inductive bias.

Main Results:

• NAGTLDA achieved high performance with an average AUC of 0.9531 and AUPR of 0.9537.
• Demonstrated superior performance compared to existing state-of-the-art methods in 5-fold cross-validation.
• Case studies validated 55 out of 60 predicted lncRNA-disease associations, highlighting the model's predictive accuracy.

Conclusions:

• The proposed NAGTLDA model offers a highly efficient computational approach for predicting biological information associations.
• The findings underscore the potential of graph Transformer structures in uncovering complex lncRNA-disease relationships.