DRMDA: deep representations-based miRNA-disease association prediction.

Area of Science:

• Genomics
• Bioinformatics
• Molecular Biology

Background:

• MicroRNAs (miRNAs) play critical roles in biological processes and are implicated in complex human diseases.
• Existing methods for predicting miRNA-disease associations have limitations.
• Accurate prediction of these associations is crucial for disease research.

Purpose of the Study:

• To develop and validate a novel computational method for predicting miRNA-disease associations.
• To address the deficiencies of previous prediction models.
• To identify potential and novel miRNA-disease relationships.

Main Methods:

• Utilized miRNA-disease association data from the HDMM database.
• Implemented a deep learning approach combining stacked auto-encoder, greedy layer-wise unsupervised pre-training, and support vector machine.
• Evaluated the DRMDA method against five established prediction models using cross-validation techniques.

Main Results:

• DRMDA achieved high performance with AUCs of 0.9177, 0.8339, and 0.9156 ± 0.0006 in global LOOCV, local LOOCV, and fivefold cross-validation, respectively.
• Case studies demonstrated high experimental validation rates for predicted miRNA-disease associations: 88% for colon neoplasms, 90% for lymphoma, and 86% for prostate neoplasms.
• The method successfully predicted top potential miRNAs for specific neoplasms.

Conclusions:

• DRMDA is a robust and promising computational tool for predicting miRNA-disease associations.
• The method can effectively identify potential and novel associations, aiding in disease research.
• DRMDA offers a significant advancement over existing prediction models.