Annotation-free prediction of microbial dioxygen utilization.

Area of Science:

• Microbiology and Genomics
• Bioinformatics and Computational Biology

Background:

• Distinguishing aerobic (oxygen-requiring) from anaerobic (non-oxygen-requiring) microbes is challenging using genomic data alone, as many microbes utilize oxygen for detoxification.
• Aerobes typically possess larger genomes with specific oxygen-utilizing enzymes, offering clues for prediction.
• Current prediction methods often rely on computationally intensive genome annotation.

Purpose of the Study:

• To develop and validate annotation-free methods for accurately predicting microbial oxygen utilization directly from genomic sequences.
• To assess the performance of sequence content-based models compared to annotation-based approaches.
• To explore the utility of these rapid prediction methods for analyzing microbial communities in natural environments.

Main Methods:

• Developed and evaluated machine learning models based on genomic sequence content, specifically amino acid trimers (triplets).
• Compared the accuracy of annotation-free models with traditional annotation-based classifiers for ternary classification of microbial oxygen utilization.
• Applied the developed models to analyze microbial community composition in the Earth Microbiome Project and a Black Sea oxygen gradient dataset.

Main Results:

• Annotation-free models using amino acid trimers achieved prediction accuracy comparable to intensive annotation-based methods (approximately 80%).
• Amino acid trimers effectively encode information about protein composition and microbial phylogeny, enabling accurate predictions.
• Analysis of the Black Sea dataset revealed a quantitative correlation between microbial community composition and local oxygen levels.

Conclusions:

• Annotation-free genomic analysis provides a rapid and accurate method for predicting microbial oxygen utilization.
• These methods can be used to infer microbial physiology and environmental conditions from DNA sequencing data.
• DNA sequencing data, analyzed with these statistical methods, can serve as a sensor for key environmental features like oxygen concentration.