GCNFORMER: graph convolutional network and transformer for predicting lncRNA-disease associations

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Disrupted expression of long non-coding RNAs (lncRNAs) is associated with various human disorders.
• Accurate prediction of lncRNA-disease associations (LDAs) is crucial for diagnostics and cost-efficiency.

Purpose of the Study:

• To develop a novel algorithm, GCNFORMER, for predicting lncRNA-disease associations.
• To leverage graph convolutional networks and transformers for enhanced LDA prediction.

Main Methods:

• Integrated intraclass similarity and interclass connections of miRNAs, lncRNAs, and diseases to build a graph adjacency matrix.
• Employed a graph convolutional network for feature extraction between nodes.
• Utilized a transformer encoder with multiheaded attention for global dependency analysis and LDA forecasting.

Main Results:

• GCNFORMER achieved high performance with AUC of 0.9739 and AUPR of 0.9812 in fivefold cross-validation.
• Demonstrated superior performance compared to six existing LDA prediction models.
• Validated effectiveness through case studies on breast, colon, and lung cancers.

Conclusions:

• The integration of graph convolutional networks and transformers significantly improves LDA prediction model performance.
• This approach facilitates advancements in lncRNA-disease association research.