Likelihood-based feature representation learning combined with neighborhood information for predicting circRNA-miRNA...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Circular RNAs (circRNAs) and microRNAs (miRNAs) play crucial roles in disease pathogenesis.
• Identifying circRNA-miRNA interactions is vital for developing diagnostic and therapeutic strategies.
• Computational models offer an efficient alternative to experimental methods for inferring these associations.

Purpose of the Study:

• To develop an efficient computational model for predicting circRNA-miRNA associations.
• To identify potential circRNA-miRNA biomarkers for disease diagnosis and treatment.
• To leverage sequence and interaction features for robust association prediction.

Main Methods:

• Developed a Convolutional Autoencoder for CircRNA-MiRNA Associations (CA-CMA) model.
• Integrated circRNA and miRNA sequence features with interaction data.
• Constructed and optimized a molecular association network using deep neural networks.
• Employed a likelihood objective to preserve neighborhood information and learn feature representations.

Main Results:

• CA-CMA achieved a mean area under the ROC curve of 0.9138 with minimal standard deviation (0.0024) during 5-fold cross-validation.
• The model demonstrated superior performance compared to existing computational approaches.
• Experimental validation confirmed the accuracy of 25 out of the top 30 predicted circRNA-miRNA pairs.

Conclusions:

• The CA-CMA model is a robust and versatile tool for predicting circRNA-miRNA associations.
• Accurate prediction of these interactions can significantly aid in disease biomarker discovery and therapeutic target identification.
• This computational approach enhances efficiency and cost-effectiveness in biomedical research.