Learning Representation of Molecules in Association Network for Predicting Intermolecular Associations

Area of Science:

• Biomedical research
• Computational biology
• Systems biology

Background:

• Post-genomic research aims to understand complex molecular interactions in human cells.
• Existing studies often focus on limited pairwise molecular associations.
• A systematic approach is needed to analyze multifaceted biomolecular networks.

Purpose of the Study:

• To develop a computational framework, MAN-SDNE, for predicting diverse intermolecular associations.
• To construct a comprehensive human molecular association network.
• To provide a tool for systematic exploration of molecular interactions and their link to diseases.

Main Methods:

• Constructed a large-scale human molecular association network integrating long non-coding RNA, microRNA, protein, drug, and disease interactions.
• Utilized network representation learning (MAN-SDNE) to capture node features based on network proximity and attributes.
• Employed a Random Forest classifier trained on these features to predict intermolecular associations.

Main Results:

• MAN-SDNE achieved high predictive performance with an AUC of 0.9552 and AUPR of 0.9338.
• Demonstrated the framework's capability through a case study on long non-coding RNA-protein interactions.
• The network contains 6,528 molecular nodes and 105,546 associations across 9 types.

Conclusions:

• MAN-SDNE offers a systematic insight into synergistic molecular associations and complex diseases.
• The developed framework provides a valuable network-based computational tool for exploring intermolecular interactions.
• This approach enhances the understanding of molecular mechanisms underlying human health and disease.