A miRNA-Disease Association Prediction Method Integrating Graph Matrix Factorization With L$_{21}$ Similarity...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• MicroRNA-disease associations are crucial for understanding disease pathogenesis and developing treatments.
• Existing graph regularized non-negative matrix factorization methods face challenges with low-dimensional matrix noise and information loss.

Purpose of the Study:

• To propose L$_{21}$ S-NPFM, a new method for miRNA-disease association prediction.
• To address limitations of existing methods by reducing noise and preserving network topology.

Main Methods:

• Introduced L$_{21}$ SGMF to incorporate a similarity constraint term, suppressing noise in low-dimensional matrices.
• Developed NPFM to fuse consistency projection matrices and initial score matrices, recovering lost network topology information.

Main Results:

• L$_{21}$ S-NPFM demonstrated superior performance compared to six other mainstream methods in LOOCV and 5-fold CV.
• Case studies achieved high accuracy: 100% for 10 miRNAs in diabetic nephropathy and 80% for 10 miRNAs in thoracic aortic aneurysm.

Conclusions:

• L$_{21}$ S-NPFM effectively predicts miRNA-disease associations by mitigating noise and integrating network information.
• The method shows significant potential for advancing disease pathogenesis research and therapeutic strategy development.