MAMLCDA: A Meta-Learning Model for Predicting circRNA-Disease Association Based on MAML Combined With CNN

Area of Science:

• Genomics and Bioinformatics
• Molecular Biology
• Computational Biology

Background:

• Circular RNAs (circRNAs) are noncoding RNA molecules with a closed, annular structure.
• Emerging evidence links circRNAs to various human diseases, highlighting the need for accurate association prediction.
• Identifying circRNA-disease associations is crucial for understanding disease mechanisms.

Purpose of the Study:

• To develop a reliable and accurate meta-learning model, MAMLCDA, for identifying circRNA-disease associations.
• To enhance the understanding of complex disease pathogenesis by exploring circRNA involvement.

Main Methods:

• A meta-learning model (MAMLCDA) combining Model-Agnostic Meta-Learning (MAML) and Convolutional Neural Network (CNN) classification was developed.
• Feature extraction and integration of circRNA-disease similarities were performed.
• K-means clustering and Probabilistic Principal Component Analysis (PPCA) were used for sample selection and feature dimensionality reduction.
• Feature vectors were converted into images for a 2-way 1-shot image classification problem.

Main Results:

• The MAMLCDA model achieved high prediction accuracies of 95.33% and 98% on two benchmark datasets.
• Cross-validation results demonstrated that MAMLCDA outperforms several existing state-of-the-art methods.
• The model effectively characterizes relationships between circRNAs and diseases.

Conclusions:

• MAMLCDA provides a robust and accurate approach for predicting circRNA-disease associations.
• The developed model can significantly contribute to elucidating the role of circRNAs in complex disease pathogenesis.
• This work advances computational methods for analyzing noncoding RNA functions in disease.