Bayesian prediction of microbial oxygen requirement.

Area of Science:

• Microbiology
• Bioinformatics
• Genomics

Background:

• Predicting bacterial habitat conditions from genome sequences is valuable for science and industry.
• Bacterial oxygen requirement is a key habitat adaptation, yet prediction from genomes is underexplored.
• Existing methods often struggle to classify more than two bacterial oxygen categories.

Purpose of the Study:

• To develop and evaluate a method for predicting bacterial oxygen requirements (aerobic, anaerobic, facultative anaerobic) using only genome sequences.
• To compare the performance of a single-step naive Bayesian classifier against a two-step Bayesian network.

Main Methods:

• Utilized genome sequence-derived features, specifically the presence or absence of protein domains.
• Employed naive Bayesian inference for classification.
• Compared a single-step prediction model with a two-step model (respiration status first, then oxygen requirement).

Main Results:

• The developed naive Bayesian method accurately predicts bacterial oxygen requirements.
• A two-step Bayesian network, first classifying respiration and then oxygen use, demonstrated superior performance.
• The method's performance is comparable to or better than existing literature methods, with increased simplicity.

Conclusions:

• A naive Bayesian network utilizing protein domain presence/absence from genome sequences is effective for predicting bacterial oxygen requirements.
• This approach offers a straightforward and efficient tool for inferring bacterial habitat preferences.