Combining non-negative matrix factorization with graph Laplacian regularization for predicting drug-miRNA...

Area of Science:

• Biomedical Informatics
• Computational Biology
• Genomics

Background:

• MicroRNAs (miRNAs) are crucial in complex human diseases.
• Targeting dysregulated miRNAs with drugs offers a novel therapeutic strategy.
• Experimental identification of drug-miRNA associations is costly and time-consuming.

Purpose of the Study:

• To develop a reliable computational method for predicting drug-miRNA associations.
• To overcome the limitations of experimental approaches in identifying drug-miRNA interactions.

Main Methods:

• Proposed GNMFDMA, combining graph Laplacian regularization with non-negative matrix factorization.
• Calculated drug and miRNA similarity matrices using biological information.
• Reformulated the drug-miRNA association matrix using nearest neighbor profiles to correct false negatives.
• Employed graph Laplacian regularization collaborative non-negative matrix factorization for association scoring.

Main Results:

• GNMFDMA achieved an AUC of 0.9193 in cross-validation, outperforming existing methods.
• Case studies validated a significant portion of the top predicted drug-miRNA associations for 5-Aza-CdR, 5-FU, and Gemcitabine.
• The method demonstrated high accuracy in identifying potential drug-miRNA relationships.

Conclusions:

• GNMFDMA is a reliable and efficient computational tool for predicting drug-miRNA associations.
• This method can accelerate the discovery of novel therapeutic strategies targeting miRNAs.
• The findings support the use of computational approaches in drug discovery and development.