MDformer: A transformer-based method for predicting miRNA-Disease associations using multi-source feature fusion and...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• MicroRNAs (miRNAs) are crucial in disease diagnosis and prognosis.
• Experimental verification of miRNA-disease associations (MDA) is inefficient.
• Existing computational methods for MDA prediction have limitations in accuracy and effectiveness.

Purpose of the Study:

• To develop an advanced computational model for predicting miRNA-disease associations (MDA).
• To improve the predictive performance and accuracy of existing computational methods.
• To leverage multi-source feature information for enhanced MDA prediction.

Main Methods:

• Proposed a transformer-based prediction model named MDformer.
• Integrated multiple miRNA and disease features from a molecular biology perspective.
• Utilized a transformer-based feature encoder and meta-path instances for node feature embeddings.
• Developed a deep neural network for MDA prediction.

Main Results:

• MDformer achieved superior performance in 5-fold cross-validation on HMDD v3.2 and HMDD v2.0 databases.
• The model demonstrated an average ROC AUC of 0.9506 and 0.9369, outperforming comparative methods.
• Case studies on five lethal cancers showed 97.3% accuracy for the top 30 predictions.

Conclusions:

• MDformer is a reliable and scientifically sound tool for accurate MDA prediction.
• The model offers a significant advancement over existing computational approaches.
• MDformer provides a valuable resource for disease research and potential therapeutic strategies.