Construction and Analysis of Molecular Association Network by Combining Behavior Representation and Node Attributes

Area of Science:

• Biomedical research
• Systems biology
• Bioinformatics

Background:

• Understanding complex biomolecular activities is crucial in post-genomic research.
• Existing studies often focus on individual molecular interactions, limiting a holistic view.
• Biomolecular interactions are interconnected and vital for sustaining life activities; disruptions can lead to diseases.

Purpose of the Study:

• To construct and analyze a large-scale human molecular association network integrating lncRNAs, miRNAs, proteins, drugs, and diseases.
• To develop a novel network representation learning method for predicting associations between molecules.
• To systematically understand and model complex biomolecular activities.

Main Methods:

• Constructed a comprehensive molecular association network (MAN) encompassing lncRNAs, miRNAs, proteins, drugs, and diseases.
• Applied network representation learning algorithm High-Order Proximity preserved Embedding (HOPE) to capture node behavior features.
• Integrated attribute features and trained a CatBoost machine learning model for predicting potential associations.

Main Results:

• Achieved high prediction accuracy of 93.3% and an area under the receiver operating characteristic curve of 0.9793 via five-fold cross-validation.
• Demonstrated the method's capability in predicting miRNA-disease associations through a case study.
• Validated the novelty of the systematic approach to intermolecular association analysis.

Conclusions:

• The developed MAN-HOPE method offers a systematic perspective on intermolecular associations.
• Enables accurate prediction of diverse molecular connections within a complex network.
• Facilitates systematic exploration of molecular interactions shaping cellular functions and disease mechanisms.