BPP: a platform for automatic biochemical pathway prediction

Area of Science:

• Biochemistry and Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Biochemical pathways are crucial for understanding life processes, involving complex networks of reactions.
• Current methods for studying these pathways, primarily experimental and database-driven, face significant cost constraints.
• Representation learning offers a promising alternative for analyzing complex biological networks.

Purpose of the Study:

• To develop an automated platform for biochemical pathway prediction (BPP).
• To identify potential links, reactants, and products within biochemical pathway networks.
• To overcome the limitations of traditional experimental and database analysis methods.

Main Methods:

• Developed the biochemical pathway prediction (BPP) platform incorporating various representation learning models.
• Utilized hypergraph neural networks to model biochemical reactions.
• Integrated novel biochemical pathway datasets and an SHAP explainer for result interpretation.

Main Results:

• The BPP platform effectively predicts potential participants and products in biochemical reactions.
• Extensive experiments demonstrated the effectiveness of the implemented models on a curated dataset.
• Case studies confirmed the platform's utility, particularly its chronological pattern analysis.

Conclusions:

• The BPP platform provides an efficient and automated solution for biochemical pathway analysis.
• Representation learning, especially hypergraph neural networks, shows great potential in modeling biochemical networks.
• The freely accessible platform, code, and datasets facilitate further research in the field.