Learning Representations to Predict Intermolecular Interactions on Large-Scale Heterogeneous Molecular Association...

Area of Science:

• Systems biology
• Computational biology
• Bioinformatics

Background:

• Molecular components in human cells form interdependent networks.
• Disturbances in molecular interactions can lead to complex diseases.
• Identifying novel biomolecular interactions is crucial for understanding regulatory mechanisms.

Purpose of the Study:

• To develop a machine learning method for predicting intermolecular interactions.
• To construct a heterogeneous molecular association network integrating various biomolecular components.
• To provide a systematic landscape for modeling complex biological associations.

Main Methods:

• Constructed a heterogeneous molecular association network including miRNAs, lncRNAs, circRNAs, mRNAs, proteins, drugs, and diseases.
• Developed a network embedding model to capture biomolecular network behavior.
• Calculated attribute features and combined them with network embeddings to train a random forest classifier (MMI-Pred).

Main Results:

• MMI-Pred achieved 93.50% accuracy in predicting hybrid associations using 5-fold cross-validation.
• The method effectively integrates diverse biomolecular data for interaction prediction.
• Demonstrated the utility of network embedding and attribute features for this task.

Conclusions:

• MMI-Pred offers a robust machine learning approach for predicting complex intermolecular interactions.
• The study provides a comprehensive framework for analyzing molecular association networks.
• This work facilitates the discovery of novel biomolecular interactions and disease mechanisms.