Predicting Multiple Types of Associations Between miRNAs and Diseases Based on Graph Regularized Weighted Tensor...

Area of Science:

• Computational biology
• Genomics
• Biomedical informatics

Background:

• MicroRNAs (miRNAs) play a crucial role in disease development through various mechanisms.
• Computational models offer efficient methods for discovering large-scale miRNA-disease associations.
• Existing models struggle with positive sample recovery and preserving biological network structures.

Purpose of the Study:

• To develop a novel computational framework, WeightTDAIGN, for identifying multiple types of miRNA-disease associations.
• To enhance the prediction performance of miRNA-disease association models.
• To address limitations in existing tensor decomposition-based models.

Main Methods:

• Proposed WeightTDAIGN, a tensor decomposition framework with weighted positive samples.
• Integrated auxiliary miRNA and disease information into the tensor decomposition.
• Applied L2,1 norm to constrain projection matrices and reduce redundant information.
• Incorporated graph Laplacian regularization to preserve biological network structure.

Main Results:

• WeightTDAIGN demonstrates improved recovery of positive samples and enhanced prediction performance.
• The model's performance is more satisfactory on sparser datasets.
• Case studies confirm WeightTDAIGN's accuracy in predicting miRNA-disease-type associations.

Conclusions:

• WeightTDAIGN is an effective computational framework for identifying miRNA-disease associations.
• The weighting strategy and graph regularization significantly improve prediction accuracy.
• The model accurately predicts complex miRNA-disease relationships, offering valuable insights for biomedical research.