Machine learning classification of archaea and bacteria identifies novel predictive genomic features

Area of Science:

• Microbiology
• Genomics
• Bioinformatics

Background:

• Archaea and Bacteria represent distinct life domains with varied ecological niches.
• Accurate classification methods exist, but underlying genomic differences require further elucidation.
• This study investigates genomic features for differentiating Archaea and Bacteria.

Purpose of the Study:

• To develop machine learning models for classifying Archaea and Bacteria using genomic features.
• To identify key genomic determinants that distinguish these two domains of life.

Main Methods:

• Utilized publicly available whole-genome sequences of bacteria and archaea.
• Extracted genomic features including nucleotide frequencies, gene content (CDS, ncRNA, rRNA, tRNA), and entropy scores.
• Developed and evaluated machine learning models (e.g., Random Forest, Neural Networks) for classification.

Main Results:

• Machine learning models achieved high classification accuracy, up to 0.998.
• Transfer RNA (tRNA) topological and Shannon's entropy emerged as the most significant discriminating genomic features.
• Nucleotide frequencies in tRNA, rRNA, ncRNA, CDS, and Chargaff's scores for tRNA and rRNA were also important.

Conclusions:

• tRNA, rRNA, and ncRNA genes are crucial genomic elements for classifying Archaea and Bacteria.
• Bacteria exhibit higher nucleotide diversity in tRNA compared to Archaea.
• Classification errors provide insights into complex phylogenetic relationships among bacteria, archaea, and eukaryotes.