DeepRF: A deep learning method for predicting metabolic pathways in organisms based on annotated genomes.

Area of Science:

• Systems Biology
• Metabolomics
• Bioinformatics

Background:

• Metabolic pathway modeling and reconstruction are crucial in systems biology due to the rise of metabolomics.
• Reconstructing an organism's metabolic network from its genome sequence is a significant challenge.
• Current methods rely on manual pathway construction, limiting scalability for large genomic datasets.

Purpose of the Study:

• To develop a scalable and accurate computational method for predicting metabolic pathways from genomic data.
• To overcome the limitations of manual pathway construction in handling large-scale sequencing data.

Main Methods:

• A supervised machine learning approach was employed, utilizing deep neural networks to learn feature representations of metabolic pathways.
• These learned features were then fed into random forests for the prediction of metabolic pathways.
• The developed model is named DeepRF.

Main Results:

• DeepRF demonstrated high performance in predicting metabolic pathways, achieving accuracy (>97%), recall (>95%), and precision (>99%) across over 318,016 instances.
• The model successfully predicts both known and unknown metabolic pathways within an organism.
• Comparative analysis indicated that DeepRF outperforms existing methods in terms of reliability.

Conclusions:

• DeepRF offers a robust and accurate solution for metabolic pathway prediction from genomic data.
• The model's high performance and scalability make it a valuable tool for systems biology and metabolomics research.
• DeepRF advances the field by enabling efficient analysis of large genomic datasets for metabolic network reconstruction.