EOESGC: predicting miRNA-disease associations based on embedding of embedding and simplified graph convolutional...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• MicroRNAs (miRNAs) are crucial in complex diseases.
• Understanding miRNA-disease associations aids pathogenesis research and drug development.
• Traditional experimental methods are time-consuming and costly.

Purpose of the Study:

• To develop an efficient deep learning model for predicting potential miRNA-disease associations.
• To overcome the limitations of traditional experimental approaches.

Main Methods:

• Constructed a coupled heterogeneous graph integrating disease similarity, miRNA similarity, and miRNA-disease associations.
• Employed an Embedding of Embedding (EOE) model to learn edge information and enrich node embeddings.
• Utilized a simplified graph convolution model for information aggregation.
• Spliced miRNA and disease feature embeddings into an MLP for prediction.

Main Results:

• The EOESGC model achieved high performance with AUC of 0.9658, AUPR of 0.8543, and F1-score of 0.8644 via 5-fold cross-validation.
• Demonstrated superior performance compared to existing state-of-the-art models.
• Successfully predicted top 20 potential miRNAs for breast and lung cancer, with high validation rates in public databases.

Conclusions:

• The EOESGC model effectively identifies potential miRNA-disease associations.
• The model offers an efficient and accurate approach for miRNA-disease association prediction.
• This work contributes to advancing disease mechanism understanding and therapeutic strategies.