An efficient circRNA-miRNA interaction prediction model by combining biological text mining and wavelet...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Circular RNAs (circRNAs) are crucial in human disease development and progression.
• Computational methods can accelerate the discovery of disease-related circRNAs.
• Existing network embedding models struggle with sparse biological networks.

Purpose of the Study:

• To develop an accurate computational model for predicting circRNA-miRNA interactions (CMI).
• To overcome limitations of existing network embedding methods in sparse biological networks.

Main Methods:

• BioDGW-CMI integrates biological text mining (BERT) with wavelet diffusion-based sparse network structure embedding.
• It constructs a circRNA-miRNA interaction network and extracts topological features.
• Denoising autoencoder and lightGBM are used for feature integration and prediction.

Main Results:

• BioDGW-CMI achieved superior performance across three CMI prediction datasets.
• The model successfully predicted all 8 tested circRNA-miRNA interactions from the circ-ITCH database.
• This indicates high accuracy and reliability in predicting functional circRNA-miRNA relationships.

Conclusions:

• BioDGW-CMI offers a robust computational approach for predicting circRNA-miRNA interactions.
• The model's effectiveness in sparse networks advances circRNA research for disease insights.
• This method holds promise for novel diagnostic and therapeutic strategies.