BoostGAPFILL: improving the fidelity of metabolic network reconstructions through integrated constraint and...

Area of Science:

• Computational Biology
• Systems Biology
• Bioinformatics

Background:

• Metabolic network reconstructions are crucial for understanding cellular metabolism but are often incomplete.
• Existing automated gap-filling tools have limitations in accuracy and validation due to experimental challenges.

Purpose of the Study:

• To develop an advanced tool, BoostGAPFILL, for metabolic model refinement and gap filling.
• To improve the accuracy and reliability of automated gap-filling methodologies.

Main Methods:

• BoostGAPFILL combines constraint-based and machine learning approaches.
• It utilizes metabolite patterns via matrix factorization to constrain reaction selection for gap filling.
• A novel testing framework was developed using existing metabolic reconstructions.

Main Results:

• BoostGAPFILL demonstrates superior performance over state-of-the-art gap-filling tools.
• The tool achieves over 60% precision and recall in predicting deleted reactions and filling gaps.
• Performance is more than double that of other existing tools across various metabolic reconstructions.

Conclusions:

• BoostGAPFILL offers a significant advancement in metabolic model refinement.
• The tool provides a more accurate and robust solution for addressing incompleteness in metabolic networks.
• Its open-source availability facilitates wider adoption and further research.