NMCMDA: neural multicategory MiRNA-disease association prediction

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• MicroRNA (miRNA) dysregulation is implicated in various diseases.
• Accurate prediction of miRNA-disease associations is vital for disease research.
• Traditional experimental methods are time-consuming and costly.

Purpose of the Study:

• To develop a novel computational method for predicting multiple-category miRNA-disease associations.
• To leverage machine learning and network analysis for improved prediction accuracy.
• To provide a time-saving and cost-effective alternative to experimental validation.

Main Methods:

• A data-driven, end-to-end learning method called Neural Multiple-Category miRNA-Disease Association prediction (NMCMDA) was developed.
• The method utilizes a Graph Neural Network encoder to learn latent representations from a miRNA-disease heterogeneous network.
• A decoder then predicts miRNA-disease association scores, with the Relational Graph Convolutional Network encoder and neural multirelational decoder (NMR-RGCN) variant showing optimal performance.

Main Results:

• The NMCMDA method, specifically the NMR-RGCN variant, demonstrated superior prediction performance compared to the state-of-the-art method TDRC.
• Performance was evaluated using Top-1 precision, recall, and F1 scores on three experimental datasets.
• Case studies on breast and lung cancer, along with validation on the HMDD v3.2 database, confirmed the method's effectiveness and feasibility.

Conclusions:

• The proposed NMCMDA method, particularly NMR-RGCN, offers a highly effective computational approach for predicting multiple-category miRNA-disease associations.
• This method significantly advances the field by providing accurate and efficient predictions.
• The findings support the utility of NMCMDA in accelerating the investigation of miRNA roles in human diseases.