A machine learning-derived genomic dataset from bacteria frequently reported as probiotics.

Area of Science:

• Microbial Genomics
• Bioinformatics
• Machine Learning

Background:

• Probiotics, live microorganisms, offer health benefits, but their effects are strain-specific.
• Identifying universal genetic markers for probiotic efficacy remains challenging due to context-dependency.

Purpose of the Study:

• To create a machine learning-derived genomic dataset of probiotic-associated features.
• To provide a resource for exploring genomic patterns in commonly used probiotic bacteria.

Main Methods:

• Comparative pangenomic analysis of bacterial genomes.
• Supervised machine learning (Random Forest, SVM, Logistic Regression) to extract discriminative genomic features.
• Generation of a dataset with protein-coding sequences, gene presence-absence matrices, and functional annotations.

Main Results:

• A dataset (ProbioSML) of 1,072 non-redundant protein-coding sequences was generated.
• Identified genomic features associated with bacterial taxa commonly used as probiotics.
• These features may reflect taxonomic and ecological signatures rather than direct causal determinants.

Conclusions:

• The ProbioSML dataset offers a valuable resource for probiogenomics and microbial genomics research.
• Publicly available data and scripts facilitate further exploratory analyses and methodological benchmarking.
• The findings highlight genomic patterns linked to probiotic bacteria, aiding future research in the field.