HOPEXGB: A Consensual Model for Predicting miRNA/lncRNA-Disease Associations Using a Heterogeneous...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Identifying disease-associated microRNAs (miRNAs) and long noncoding RNAs (lncRNAs) is vital for developing biomarkers for complex human diseases.
• Computational prediction of miRNA/lncRNA-disease associations offers a cost-effective and time-efficient alternative to experimental methods.

Purpose of the Study:

• To develop a novel computational approach, HOPEXGB, for predicting disease-related miRNAs and lncRNAs.
• To integrate diverse biological information, including similarities, correlations, and interactions, within a heterogeneous network for improved prediction accuracy.

Main Methods:

• A heterogeneous disease-miRNA-lncRNA (DML) information network was constructed by linking nodes based on their relationships.
• High-order proximity preserved embedding (HOPE) and eXtreme Gradient Boosting (XGB) were employed in a consensual machine-learning framework.
• A refined negative dataset generation strategy was implemented to minimize false negatives.

Main Results:

• The HOPEXGB model achieved a mean prediction accuracy of 0.9569, outperforming existing methods.
• HOPE demonstrated superior performance compared to other graph embedding techniques in 10-fold cross-validation.
• The model exhibited high sensitivity and specificity, with promising results on external validation datasets.

Conclusions:

• Integrating lncRNA-miRNA interactions and similarity information significantly improves prediction performance.
• HOPEXGB serves as a powerful tool for preclinical biomarker detection and preliminary screening for cancer diagnosis and prognosis.
• The HOPEXGB tool is publicly available for broader research application.