Hessian Regularized -Nonnegative Matrix Factorization and Deep Learning for miRNA-Disease Associations Prediction

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• MicroRNAs (miRNAs) are critical regulators in biological processes, and their dysregulation is linked to various human diseases.
• Experimental identification of miRNA-disease associations is resource-intensive and time-consuming.
• Computational prediction of these associations offers valuable preliminary insights for researchers.

Purpose of the Study:

• To develop a novel computational model for predicting potential miRNA-disease associations.
• To improve the accuracy and efficiency of miRNA-disease association prediction compared to existing methods.
• To provide a tool that aids in the early diagnosis and treatment of complex human diseases.

Main Methods:

• Developed a hybrid model, Hessian-regularized nonnegative matrix factorization with deep learning (H-NMF-DF).
• Employed an iterative fusion approach to integrate multiple similarity matrices, reducing data sparsity.
• Utilized a mixed model framework combining deep learning, matrix decomposition, and singular value decomposition to capture nonlinear features.

Main Results:

• The H-NMF-DF model demonstrated competitive or superior prediction performance (AUC and AUPR) compared to six other matrix factorization methods.
• Case studies on lung, bladder, and breast tumors validated the model's high accuracy in predicting disease-related miRNAs.
• The hybrid approach effectively addresses data sparsity and captures complex biological interactions.

Conclusions:

• The proposed H-NMF-DF model accurately predicts miRNA-disease associations, offering a valuable tool for biomedical research.
• This computational approach can accelerate the discovery of novel diagnostic and therapeutic targets for complex diseases.
• The integration of matrix factorization and deep learning presents a powerful strategy for biological data analysis.