HoRDA: Learning higher-order structure information for predicting RNA-disease associations

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Circular RNAs (CircRNAs) and microRNAs (miRNAs) are key non-coding RNAs implicated in diseases.
• Predicting RNA-disease associations is vital but challenging due to complexity and cost.
• Existing deep learning methods lack universal accuracy and fail to capture higher-order topological information.

Purpose of the Study:

• To develop a novel deep learning framework, HoRDA, for accurate RNA-disease association prediction.
• To leverage higher-order structure information in biological networks for improved prediction.
• To address limitations of existing methods in universality and accuracy.

Main Methods:

• Utilized a higher-order graph attention network to explore correlations between RNAs and diseases.
• Employed a higher-order graph convolutional network to aggregate neighbor information and derive RNA/disease representations.
• Implemented a higher-order negative sampling strategy to generate effective negative samples.
• Integrated RNA and disease embeddings into a logistic regression model for probability prediction.

Main Results:

• HoRDA demonstrated superior performance compared to existing methods in diverse simulations.
• The method effectively captures higher-order topological information crucial for RNA-disease associations.
• Case studies on breast, colorectal, and gastric neoplasms validated the practical applicability and effectiveness of HoRDA.

Conclusions:

• HoRDA offers a significant advancement in predicting RNA-disease associations by incorporating higher-order network structures.
• The proposed higher-order strategies enhance prediction accuracy and universality.
• HoRDA provides a valuable tool for biological research and disease association studies.