A learning based framework for diverse biomolecule relationship prediction in molecular association network.

Area of Science:

• Computational biology
• Systems biology
• Bioinformatics

Background:

• Cellular functions rely on complex biomolecule interactions.
• Previous models often overlook the holistic nature of cells, focusing on isolated components.
• Understanding these relationships is crucial for deciphering cellular regulatory mechanisms.

Purpose of the Study:

• To develop a computational model that captures the holistic nature of cellular biomolecule associations.
• To predict relationships among different biomolecule types within a cell.
• To provide a novel perspective for elucidating cellular regulatory mechanisms.

Main Methods:

• Construction of a Molecular Associations Network (MAN) encompassing 9 relationship types among 5 biomolecule types.
• Development of the biomarker2vec algorithm to represent biomolecules as vectors, integrating attribute (k-mer) and behavior (Graph Factorization) information.
• Application of a Random Forest classifier for model training, validation, and testing.

Main Results:

• The MAN-GF model achieved high performance with an Area Under the Curve (AUC) of 0.9647 and Area Under the Precision-Recall Curve (AUPR) of 0.9521.
• The model demonstrated substantial predictive accuracy under 5-fold cross-validation.
• The holistic approach of MAN-GF proved effective in capturing complex biomolecular interactions.

Conclusions:

• The MAN-GF model, by adopting an overall perspective of cellular networks, serves as a valuable tool for practical applications.
• This approach offers a promising new avenue for understanding intricate cellular regulatory mechanisms.
• The study highlights the importance of holistic modeling in computational biology.