Predicting multiple types of miRNA-disease associations using adaptive weighted nonnegative tensor factorization with...

Area of Science:

• Biomedical Informatics
• Computational Biology
• Genomics

Background:

• MicroRNA (miRNA) dysregulation is linked to various diseases.
• Identifying disease-related miRNAs is crucial for understanding disease mechanisms.
• Computational models offer efficient alternatives to traditional experiments for miRNA-disease association discovery.

Purpose of the Study:

• To develop an advanced computational framework for discovering potential multiple types of miRNA-disease associations.
• To address key challenges in existing tensor-based models, including local minima, high-order relation preservation, and false-negative samples.

Main Methods:

• Proposed a novel tensor completion framework: SPLDHyperAWNTF (Self-Paced Learning, Hypergraph Regularization, Adaptive Weight Tensor, Nonnegative Tensor Factorization).
• Integrated self-paced learning with nonnegative tensor factorization to avoid local minima.
• Employed hypergraph regularization to preserve high-order relationships among miRNAs and diseases.
• Introduced an adaptive weight tensor to mitigate the impact of false-negative samples.

Main Results:

• SPLDHyperAWNTF demonstrated superior prediction performance over baseline models in cross-validation studies (Top-1 precision, recall, F1).
• Case studies confirmed a high percentage of predicted associations (98 out of 100) against the HMDDv3.2 dataset.
• Enrichment analysis indicated biological significance even for unconfirmed potential associations.

Conclusions:

• SPLDHyperAWNTF effectively identifies potential miRNA-disease associations, overcoming limitations of previous computational models.
• The proposed framework offers a robust and accurate method for advancing miRNA-disease association research.
• The findings contribute to a deeper understanding of miRNA roles in disease pathogenesis.