HTFSMMA: Higher-Order Topological Guided Small Molecule-MicroRNA Associations Prediction.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Small molecules (SMs) are crucial regulators of microRNAs (miRNAs).
• Current SM-miRNA association prediction methods often neglect local topological features and effective feature fusion.
• Accurate prediction of SM-miRNA interactions is vital for drug discovery and biological research.

Purpose of the Study:

• To develop a novel computational approach, HTFSMMA, for predicting associations between small molecules and microRNAs.
• To effectively incorporate local topological features and fuse node and topological features for improved prediction accuracy.
• To address limitations in existing methods for SM-miRNA association prediction.

Main Methods:

• Constructed an association graph integrating SM-miRNA data and node similarities.
• Utilized a target neighborhood graph convolutional network to extract local topological features.
• Employed graph attention networks and random walks to acquire high-order features.
• Integrated extracted features into a multilayer perceptron for association score prediction.

Main Results:

• HTFSMMA demonstrated superior performance across multiple cross-validation strategies.
• Area under the receiver operating characteristic curve values consistently exceeded 0.99 in rigorous evaluations.
• Case studies and statistical tests confirmed the method's effectiveness and reliability.

Conclusions:

• HTFSMMA significantly advances the prediction of small molecule-microRNA associations.
• The method's ability to integrate diverse features offers a robust framework for biological network inference.
• HTFSMMA holds promise for accelerating the identification of novel therapeutic targets and drug candidates.