Metabolic pathway inference using multi-label classification with rich pathway features.

Area of Science:

• Systems Biology
• Genomics
• Computational Biology

Background:

• Metabolic inference from genomic data is crucial for understanding cellular capabilities.
• Current gene-centric methods have limitations in predicting pathway presence and lack standardization.
• Pathway-centric approaches using machine learning offer improved hypothesis generation for metabolic relationships.

Purpose of the Study:

• To introduce mlLGPR, a novel software package for metabolic network inference.
• To utilize supervised multi-label classification with rich pathway features for enhanced prediction.
• To infer metabolic networks in both single-organism and multi-organismal datasets.

Main Methods:

• Developed mlLGPR (multi-label based on logistic regression for pathway prediction).
• Employed supervised multi-label classification algorithms.
• Utilized rich pathway features for training and prediction.

Main Results:

• Evaluated mlLGPR on 12 diverse experimental datasets, including genomes and microbial communities.
• Achieved performance metrics equal to or exceeding previous methods for organismal genomes.
• Identified challenges in feature engineering and training data for community-level inference.

Conclusions:

• mlLGPR provides a robust method for metabolic network inference.
• The tool demonstrates strong performance for organismal genomes and offers potential for community analysis.
• Further research is needed to optimize feature engineering and training data for microbial community metabolic inference.