Motif-Aware miRNA-Disease Association Prediction via Hierarchical Attention Network

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Micro-ribonucleic acids (miRNAs) function as post-transcriptional regulators and are potential biomarkers for various diseases.
• Predicting miRNA-disease associations (MDAs) is crucial for understanding disease mechanisms and progression.
• Current MDA prediction models often overlook the utility of network motif information.

Purpose of the Study:

• To develop a novel computational model, MotifMDA, for predicting miRNA-disease associations.
• To leverage high- and low-order structural information within the miRNA-disease association network.
• To incorporate motif-level network patterns for improved prediction accuracy.

Main Methods:

• Designed specific network motifs to capture diverse miRNA-disease association patterns.
• Employed a two-layer hierarchical attention mechanism within the MotifMDA model.
• Learned high-order motif preferences and coupled them with low-order preferences for final miRNA and disease embeddings.

Main Results:

• MotifMDA demonstrated superior performance compared to state-of-the-art models on two benchmark datasets.
• Accurate miRNA-disease association prediction was achieved using solely network information.
• Case studies confirmed MotifMDA's ability to discover novel MDAs from different structural perspectives.

Conclusions:

• Network motif information is highly valuable for accurate miRNA-disease association prediction.
• The MotifMDA model offers a powerful approach for identifying potential disease biomarkers.
• Integrating motif-level structural patterns enhances the discovery of novel miRNA-disease relationships.