LDAformer: predicting lncRNA-disease associations based on topological feature extraction and Transformer encoder

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Identifying long noncoding RNA (lncRNA)-disease associations is crucial for disease diagnosis and treatment.
• Current computational methods for predicting these associations often have limitations in feature extraction and model complexity.

Purpose of the Study:

• To propose a novel method, LDAformer, for predicting lncRNA-disease associations.
• To improve predictive performance by enhancing feature extraction and employing a powerful learning model.

Main Methods:

• Constructed a heterogeneous network integrating lncRNA, disease, and microRNA (miRNA) associations.
• Employed topological feature extraction to capture multi-hop pathway features from a weighted adjacency matrix.
• Utilized a Transformer encoder with global self-attention for predicting lncRNA-disease associations.

Main Results:

• LDAformer demonstrated superior performance compared to state-of-the-art baseline methods on two datasets.
• The method achieved ideal performance due to efficient feature extraction and an intuitive learning model.
• Case studies confirmed LDAformer's accuracy in discovering novel lncRNA-disease associations.

Conclusions:

• LDAformer offers an effective approach for predicting lncRNA-disease associations.
• The method's advanced feature extraction and Transformer-based model contribute to its high predictive power.
• LDAformer shows significant potential for advancing disease diagnosis and treatment strategies through accurate association discovery.