An efficient model for predicting human diseases through miRNA based on multiple-types of contrastive learning

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• MicroRNAs (miRNAs) play a crucial role in regulating human microbiota and disease development.
• Accurate prediction of miRNA-disease associations (MDAs) is vital for early diagnosis and treatment.
• Existing machine and deep learning models for MDAs struggle with data sparsity.

Purpose of the Study:

• To develop an advanced model for predicting potential miRNA-disease associations (MDAs).
• To overcome the challenge of sparse node neighborhoods in existing prediction methods.
• To enhance the accuracy and reliability of MDA prediction for clinical applications.

Main Methods:

• Proposed a novel model, MTCL-MDA, integrating multiple contrastive learning strategies with graph collaborative filtering.
• Employed a topology-based contrastive learning strategy to mitigate performance degradation from sparse neighborhoods.
• Utilized a semantic-based contrastive learning strategy to reduce noise and enrich node semantic information.

Main Results:

• The MTCL-MDA model demonstrated superior performance across all evaluation metrics compared to existing methods.
• Case analysis confirmed the model's enhanced accuracy in identifying potential MDAs.
• The findings suggest significant implications for disease screening and diagnosis.

Conclusions:

• MTCL-MDA effectively addresses data sparsity in miRNA-disease association prediction.
• The model's superior performance and accuracy offer a valuable tool for clinical applications.
• Publicly available data and code facilitate further research and development in this area.