Predicting miRNA-Disease Associations by Incorporating Projections in Low-Dimensional Space and Local Topological...

Area of Science:

• Bioinformatics
• Genomics
• Computational Biology

Background:

• Identifying microRNA (miRNA) associations with diseases aids in understanding disease mechanisms.
• Current methods leverage miRNA/disease similarities and known associations but overlook low-dimensional projections.
• A need exists for methods incorporating low-dimensional feature representations for improved miRNA-disease prediction.

Purpose of the Study:

• To develop a novel computational method for predicting potential miRNA-disease associations.
• To integrate low-dimensional feature representations of miRNAs and diseases into a predictive model.
• To enhance the accuracy and credibility of miRNA-disease association predictions.

Main Methods:

• Proposed DMAPred, a method based on non-negative matrix factorization.
• Exploited miRNA/disease similarities, known associations, and miRNA network topology.
• Projected miRNAs and diseases into a low-dimensional feature space for dense representations.
• Incorporated the sparse nature of miRNA-disease associations.

Main Results:

• DMAPred demonstrated superior performance across 15 well-characterized diseases.
• Achieved high Area Under the Receiver Operating Characteristic Curve (AUC) values (0.860–0.973).
• Achieved significant Area Under the Precision-Recall Curve (AUPR) values (0.118–0.761).
• Case studies on breast, prostate, and lung neoplasms validated DMAPred's predictive capability.

Conclusions:

• DMAPred effectively predicts potential miRNA-disease associations by utilizing low-dimensional projections and network information.
• The method offers a credible and accurate approach for discovering novel disease-related miRNAs.
• This work contributes to advancing the understanding of miRNA roles in disease pathogenesis.