Metabolic Pathway Prediction Using Non-Negative Matrix Factorization with Improved Precision.

Area of Science:

• Computational Biology
• Bioinformatics
• Systems Biology

Background:

• Machine learning offers probabilistic metabolic pathway inference from genomic data.
• Challenges include feature engineering, reaction mapping, and distributed metabolism, limiting prediction performance.

Purpose of the Study:

• To introduce triUMPF (triple non-negative matrix factorization with community detection) for enhanced metabolic pathway inference.
• To address limitations in current prediction methods for complex biological systems.

Main Methods:

• Utilizes triple non-negative matrix factorization (NMF) to model enzyme-pathway relationships in a graph network.
• Employs community detection to identify higher-order structures based on vertex clustering.
• Evaluated on diverse datasets including BioCyc Tier 1 genomes and microbial communities.

Main Results:

• triUMPF achieved performance metrics equal to or exceeding existing prediction methods for organismal genomes.
• Demonstrated improved precision in predicting metabolic pathways for multi-organismal datasets.
• Successfully captured complex relationships in both single genomes and microbial communities.

Conclusions:

• triUMPF offers a robust framework for metabolic pathway inference, outperforming previous methods.
• The integration of NMF and community detection enhances the accuracy and precision of predictions.
• This method is valuable for analyzing metabolic functions across various biological scales.