Decoding oxygen preference: Machine learning discovers functional genes in Bacteria.

Area of Science:

• Microbiology
• Genomics
• Bioinformatics

Background:

• Predicting bacterial oxygen preference is crucial for understanding microbial physiology and ecology.
• Identifying genes involved in oxygen adaptation is essential for both basic research and applied microbiology.

Purpose of the Study:

• To develop and validate a machine learning model for predicting bacterial oxygen preference using genomic data.
• To identify novel genes and protein domains associated with bacterial oxygen adaptation.
• To apply the developed model to analyze microbial communities, such as those found in rumen metagenomes.

Main Methods:

• A Random Forest machine learning model was trained on genomic features from 1813 bacterial genomes.
• Feature importance analysis was performed to identify key genomic determinants of oxygen preference.
• Experimental validation involved gene overexpression in Escherichia coli to assess functional roles in oxygen adaptation.

Main Results:

• The Random Forest model achieved 90.62% accuracy in predicting bacterial oxygen preference, surpassing previous methods.
• Key protein domains (SOD, SAM radical enzyme, GCV-T, FDH) and candidate genes were identified as significant predictors.
• Overexpression of identified genes enhanced aerobic growth in E. coli, confirming their role in oxygen adaptation.
• Application to rumen metagenomes indicated a predominantly anaerobic microbial community.

Conclusions:

• Machine learning provides an effective strategy for predicting bacterial oxygen preference and discovering functional genes.
• This approach offers a novel tool for understanding bacterial oxygen adaptation mechanisms.
• The study facilitates the exploration of uncultured microbial resources and their metabolic potential.